Sie sind auf Seite 1von 19

j o u r n a l o f t r a f fi c a n d t r a n s p o r t a t i o n e n g i n e e r i n g ( e n g l i s h e d i t i o n ) 2 0 1 7 ; 4 ( 6 ) : 5 0 3 e5 2 1

Available online at www.sciencedirect.com

ScienceDirect

journal homepage: www.elsevier.com/locate/jtte

Original Research Paper

Development of a bicycle level of service model for


urban street segments in mid-sized cities carrying
heterogeneous traffic: A functional networks approach

Sambit Kumar Beura*, Prasanta Kumar Bhuyan


Department of Civil Engineering, National Institute of Technology Rourkela, Rourkela 769008, India

highlights

 A bicycle level of service (BLOS) model is developed for the mixed traffic flow conditions.
 Functional networks (FNs) approach is implemented to predict the BLOS with high efficiency.
 Outermost traffic lane has the highest influence on quality of bicycling.
 Interruption by intermittent public transits is one newly introduced variable.
 Majority of Indian roads are offering average to very poor BLOS in the present scenario.

article info abstract

Article history: Roadways in developing countries usually carry heterogeneous traffic where on-street bi-
Received 25 November 2016 cyclists encounter a very complex interaction with various categories of vehicles. In order
Received in revised form to quantify the operational conditions of bicyclists under such conditions, a reliable bicycle
8 February 2017 level of service (BLOS) model is yet to be devised. This study intends to partially fill this gap
Accepted 17 February 2017 by proposing a BLOS model suitable for urban road segments in mid-sized cities carrying
Available online 15 November 2017 heterogeneous traffic. A recently introduced artificial intelligence technique namely,
associativity functional network (FN) is implemented for the development of this model.
Keywords: FN is a problem-driven approach that overcomes several limitations of the artificial neural
Bicycle level of service network (ANN) technique. The urban bicycling environments persisting on 74 street seg-
Urban road segment ments are analyzed, and 8 influencing variables (geometric, traffic and built environ-
Mid-sized city mental, etc.) are identified. Of these variables, interruptions caused by frequent stoppages
Functional network of intermittent public transits and frequency of driveways carrying high volume of traffic
Sensitivity analysis are newly introduced. In the modeling process, a strong relationship has been established
Developing country between the identified variables and perceived BLOS scores collected through perception
surveys. The resulting BLOS model has shown a high reliability for its applications in the
mid-sized cities and has reported a high correlation coefficient (R) of 0.94 with the average
observations. Besides, a sensitivity analysis is also carried out to identify the relative
importance of input variables based on their contribution in the BLOS estimation. As
observed, effective width of the outside lane, traffic volume, and on-street parking activity
are by far the most important variables, which contribute 38.3%, 21.8%, and 12.7%

* Corresponding author. Tel.: þ91 9938250999.


E-mail addresses: sambit.beura@gmail.com (S. K. Beura), pkbtrans@gmail.com (P. K. Bhuyan).
Peer review under responsibility of Periodical Offices of Chang'an University.
https://doi.org/10.1016/j.jtte.2017.02.003
2095-7564/© 2017 Periodical Offices of Chang'an University. Publishing services by Elsevier B.V. on behalf of Owner. This is an open
access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
504 J. Traffic Transp. Eng. (Engl. Ed.) 2017; 4 (6): 503e521

respectively in the prediction of facility BLOS. Thus, these three attributes should be largely
prioritized while making any plan of actions for the betterment of bicyclists.
© 2017 Periodical Offices of Chang'an University. Publishing services by Elsevier B.V. on
behalf of Owner. This is an open access article under the CC BY-NC-ND license (http://
creativecommons.org/licenses/by-nc-nd/4.0/).

any transportation system. Empirical models built upon


1. Introduction important road attributes help to identify the service levels at
which existing roadways are accommodating bicycle users.
An analysis of the survey data from 150 countries reports that Such models are traditionally acknowledged as bicycle level of
around 42% of households on the planet are owned with at least service (BLOS) models. The service quality assessment quan-
one bicycle (Oke et al., 2015). In addition to seamless door-to- tifies how much there is a need to redesign the existing fa-
door mobility, several economic, environmental, health and cilities for making them abundantly bicycle friendly. Outputs
societal benefits have made the bicycle mode highly desirable from BLOS models also help to identify which attributes of the
for short trips. According to Census 2011, as high as 44.8% of roadways should be primarily prioritized for the augmenta-
the households in India, a developing country with a tion of service quality. As yet, several researchers in devel-
population size of above 1.3 billion, own bicycle(s). This oped countries have carried out in-depth research works and
implies that a significant percentage of the total population in proposed several BLOS models. However, those models will
India use this mode. However, the bicycle mode share in the fail to fulfill the safety needs and mobility requirement of bi-
country is gradually declining amid rising in the wealth level cycle users in developing countries due to differing situations.
of citizens. Bicycle-sharing system facility which ensures the Roadways in developed countries generally carry homoge-
public freedom to use some specific bicycles and leave them as neous traffic where road users follow the lane discipline. On
necessary (Abagnale et al., 2014a,b), is also hardly available in the contrary, road infrastructures in developing countries
India and other developing countries at present. The road usually carry heterogeneous traffic where motorized and non-
infrastructure is designed for a mixed traffic flow of well- motorized 2-wheerers, 3-wheelers and 4-wheelers, etc., share
protected fast moving motorized vehicles with unprotected the available road space with a lack of lane discipline. Thus,
slow moving bicycles. Hence, when efforts are made to on-street bicyclists here encounter a very complex interaction
accommodate bicyclists along with motorists within the same with various categories of vehicles which largely vary in their
road space, any uncertain errors or unpredictable actions of sizes and operating conditions. In addition to this complex
the users often lead to crashes. Moreover, an annual growth situation, the geometric designs of roadways in developing
rate of around 11% for motorized two-wheelers and cars, and countries are also significantly different from the same in
7%e8% for other motorized modes have also been degrading developed countries. As a result, the operational characteris-
the bicyclists' sense of safety under such a contained tics of bicyclists under heterogeneous traffic situations
transportation environment. significantly differ from that under homogeneous traffic
The high transportation activities, particularly at city situations.
centers, are also contributing to high air polluting emissions The primary objective of this study is to develop a new
and thus greatly degrading the quality of bicycling environ- BLOS model for urban street segments in mid-sized cities
ment. Though the level of emissions from traffic have been (population range 0.5e1.0 million) operating under heteroge-
reduced with technical developments in the fields of engine neous traffic flow conditions. In order to accomplish this, the
and after-treatment technologies, many urban areas around bicycling environment persisting on 74 urban road segments
the world still continue to suffer from high concentrations of situated in four different Indian mid-sized cities are thor-
these emissions. Mopeds and motorcycles together play a vital oughly analyzed. These segments widely vary from each other
role in transportation activities occurring in developing with respect to their geometrical designs, traffic flow condi-
countries, particularly in southern European and eastern as tions, and the surrounding environment, and well represent
well as southern Asian countries (Iodice et al., 2010; Iodice and the variability and complexities in the prevailing roadway
Senatore, 2012) and emits significant amount of pollutants to environments. Eight different attributes of the roadways
the air which, on the other hand, hugely affect unprotected having significant effect on the quality of bicycling environ-
bicyclists. Mopeds (Euro II Emission standard) and motorcy- ment are identified, such as effective width of the outermost
cles (< 250 cm3, Euro III Emission Standard), for instance, have through lane, peak hour traffic volume per lane, average
emitted a total of 2186.3 and 350.5 t/year of carbon monoxide traffic speed, pavement condition index, roadside commercial
(CO), 450.8 and 64.6 t/year of volatile organic compounds activities, interruptions by unauthorized stoppages of inter-
(VOC), and 111.7 and 31.5 t/year of nitrogen oxides (NOx) mittent public transits, vehicular ingresseegress to the on-
respectively in Campania region (Italy) in 2011 (Iodice and street parking area, frequency of driveways carrying high
Senatore, 2013). volume of traffic. Subsequently, a BLOS model is developed by
The modeling of bicycle service qualities is imperatively an incorporating these important variables as independent var-
important issue for the planning, design, and management of iables and overall perceived satisfaction scores of bicyclists
J. Traffic Transp. Eng. (Engl. Ed.) 2017; 4 (6): 503e521 505

(assessed through a stated preference survey) as the output hazard score (IHS), proposed for similar resolutions, signifies
variable. the importance of roadside land use intensity and curb cut
Though regression analysis has found wide applications in frequency (or on-street parking frequency) in determining the
the literature for BLOS as well as pedestrian level of service bicycle service quality (Landis, 1994). Bicycle stress level (BSL)
(PLOS) models development, the applicability and perfor- model reflects the prominence of curb-lane, traffic volume
mance of an advantageous and recently introduced artificial and traffic speed in quantifying the riding quality of bicyclists
intelligence technique namely, functional networks (FNs) has (Sorton and Walsh, 1994). Another service measuring model
been investigated in this study. The architecture of functional namely, bicycle suitability score (BSS) model assesses the
networks approach is based upon the structure of the physical BLOS of urban road segments by using shoulder or travel lane
world. It overcomes several limitations of the artificial neural width, average daily traffic, average traffic speed and
networks (ANNs) approach. pavement surface quality parameters (Turner et al., 1997).
Goodness-of-fit and prediction precision of the FNs based Various studies have exposed the important roles of bicycle
BLOS model proposed in this study are assessed using several lane parameters and pavement surface quality in
statistical parameters such as: correlation coefficient (R), determining the BLOS of urban roads (Chellapilla et al., 2016;
NasheSutcliffe efficiency coefficient (E), error measuring pa- FDOT, 2009; Harkey et al., 1998; Jensen, 2007; Landis et al.,
rameters (average absolute error (AAE), maximum absolute 1997). The developers of bicycle compatibility index (BCI)
error (MAE), and root mean square error (RMSE)) and over- model have exposed the importance of bicycle lane
fitting ratio (OR), etc. In addition to this effort, a sensitivity parameters and right turning vehicles in BLOS estimation
analysis is also carried out to identify the percentage contri- (Harkey et al., 1998).
bution of each modeled parameter in predicting the bicycle Few other studies have investigated the benefits of con-
service quality. The remaining parts of this manuscript are verting a wide outermost lane into a bicycle lane and a traffic
organized as follows: discussion on background studies, study lane (Hallett et al., 2006; Hunter et al., 2005). The cited authors
methodology, site selection and data collection, data analyses, have concluded that bicyclists gain better confidence to ride
model development and validation, model applications, further from the edge of roads (and consequently closer to the
findings, and conclusions. motorized vehicles) if there is a provision of separate bicycle
lane. Thus, the provision of a bicycle lane is perhaps more
desirable than a wide outside lane in order to accommodate
2. Background of the study the bicycle traffic in a safer and better way. BLOS criteria
proposed in 2000 version of Highway Capacity Manual (TRB,
An in-depth review of the background studies is carried out to 2000) is based on average traffic speed, average delay, and
gain an insight into the state-of-the-art of the focused area of hindrance. However, 2010 version of Highway Capacity
research. The review findings are summarized under two Manual (TRB, 2010) considers a wide range of factors such as
major sections: (1) discussions on existing bicycle level of effective width of the outermost lane, mid-segment demand
service (BLOS) models, (2) exclusive summary of modeling flow rate, number of lanes, traffic speed, percentage of
approaches which are being followed in this particular area of heavy vehicles and pavement surface quality. A BLOS model
research, and the evolutionary history of the advantageous developed from users' perspective has revealed that the
technique adopted in this study, i.e., functional networks BLOS is largely determined by width of the road on which
(FNs). bicycle is ridden (Kang and Lee, 2012). Other significant
parameters considered in this model are number of lanes,
2.1. Existing bicycle level of service (BLOS) model pedestrian volume and number of encounters. An
investigation of the factors influencing bicycling in an Indian
At present, several well-established BLOS models are available city namely, Bangalore concludes that segregated cycle lanes
for the assessment of service quality offered by road segments and signals at intersections are two essential requirements
carrying homogeneous traffic. However, none of them is suit- of a safe bicycling environment (Verma et al., 2016). Recent
able for roadways carrying heterogeneous traffic. Among studies on bicycle operations under heterogeneous traffic
existing BLOS models, the development of the most initial flow conditions have concluded that the quality of bicycling
model namely roadway segment index (RSI) model was is largely influenced by vehicular traffic volume (Beura et al.,
accomplished by Davis (1987). This model assesses the quality 2016; Chellapilla et al., 2016). On-road bicyclists under such
of road segments by using the information on number of conditions encounter a very complex interaction with
through traffic lanes, width of the outermost lane, traffic various categories of small to big vehicles and subsequently
volume, speed limit, pavement factors and location factors. have their quality of riding largely influenced.
Modified roadway condition index (modified RCI) model is As indicated in the earlier discussions, the applicability of
one of the revised versions of RSI model in which authors existing BLOS models built upon homogeneous traffic flow
have modified location and pavement factors, and multiplied situations is suspicious in a developing country like India
the lane width parameter with speed limit to place larger where the traffic flow is highly heterogeneous and road
weightage on narrow roads with high traffic speeds geometric conditions are quite different. In addition, existing
(Epperson, 1994). Another modified versions of RSI model, models do not address the influence of few other factors
known as bicycle suitability rating (BSR) model, signifies the namely, unauthorized stoppages of intermittent public tran-
important roles of traffic volume and traffic speed in sits and driveways carrying a high volume of traffic.
operational conditions of bicyclists (Davis, 1995). Interaction Researchers neglected these variables have probably
506 J. Traffic Transp. Eng. (Engl. Ed.) 2017; 4 (6): 503e521

assumed (or observed) these variables to be insignificant in


the BLOS models development under homogeneous traffic 3. Functional networks (FNs) modeling
conditions. However, these variables have different roles to procedure
play in BLOS estimation in developing countries. In view of
all these aspects, the present study has carried out a thor- FN, a powerful alternative to the ANN technique, deals with
ough investigation to identify all significant variables. generalized functional models instead of standard types
Further, an empirical BLOS model is proposed which is (Castillo, 1998; Castillo and Ruiz-Cobo, 1992; Castillo et al.,
basically new decision support system for transport planners 1998, 2000a, 2000b). Its architecture is based on the structure
and engineers. of the physical world. The neuron functions associated with
each neuron of the FNs architecture are not fixed but are
learned entirely from the data at hand. The initial architecture
of FNs for any problem is designed based on the type and
2.2. Modeling approach
characteristics of that particular problem. Thus, FN is often
called as a problem-driven method. Besides, FNs technique
All of the existing BLOS models discussed earlier are devel-
also has the ability to take into account the domain knowl-
oped using statistical techniques such as regression analysis,
edge, i.e., properties of the function being modeled such as
point system, logit modeling, or probit modeling approach.
associative, commutative and distributive, while establishing
However, in terms of prediction precision, artificial intelli-
the relationship between input and output variables. Using
gence (AI) techniques are gaining increasing popularity over
the domain knowledge, FNs make assumptions about the
these statistical techniques in various fields of engineering
form that unknown functions should take. As FNs use both
and technology. This is why, the present study has taken the
the domain knowledge and the data knowledge to derive the
help of a recently developed artificial intelligence technique
topology of a problem, they eliminate the problems of ANNs
namely, functional networks (FNs) for the development of the
being “black boxes”. Another advantage of the FNs technique
new BLOS model. FN architecture, proposed and promoted by
is that it allows the neurons to be multi-argument, multivar-
several researchers (Castillo, 1998; Castillo and Ruiz-Cobo,
iate, and different learnable functions, instead of any fixed
1992; Castillo et al., 1998, 2000a, 2000b), is one of the advanced
functions. Following subsections give a brief description on (a)
prediction tools. It has found its efficient application in the
main components of functional networks, (b) the working
literature in solving several prediction and classification
principles of the functional networks.
problems. Several added advantages have made it as a
powerful alternative to the artificial neural network (ANN)
technique. FNs are being considered as a general framework 3.1. Main components of functional networks (FNs)
suitable for solving a wide range of problems in probability,
signal processing, pattern recognition, classification, statis- Fig. 1 displays the main elements usually encountered in each
tics, systems identification, real-time forecasting, functions functional network.
approximations, bioinformatics applications, and other busi- A brief discussion on each major component of a func-
ness and engineering applications. The huge reliability and tional network is given below.
prediction precision of FNs technique in all these areas of
research has largely motivated the present study to test its  Several layers of storing units. The storing units store in-
performance in predicting the bicycle service quality. The puts, output, and other intermediate information of the
following section gives a brief discussion on advantages, dis- functional network. In the functional network shown in
advantages and the underlying principles of the FNs modeling Fig. 1(a), the small filled circles represent the storing units.
procedure. There are three layers of storing units: the first (leftmost)

Fig. 1 e Main elements in functional networks. (a) A typical functional network. (b) The equivalent simplified associativity
functional network.
J. Traffic Transp. Eng. (Engl. Ed.) 2017; 4 (6): 503e521 507

layer contains three input units x1, x2, and x3, the second validation process is satisfactory, the model is ready for
layer (middle) contains four intermediate units and the implementation.
third (rightmost) consists of one output unit (x0).
 One or more layers of neurons (computing or functional This study has used the most recent and probably the
units). The open circles in Fig. 1 represent the neurons of simplest form of the functional networks called as “associa-
the functional network. Each neuron is a computing unit tivity functional networks” for the development of BLOS model.
which assesses a set of inputs and returns a set of A brief discussion on underlying principles of associativity
outputs to the next layer of storing units. Hence, each functional networks is presented below for completeness.
neuron represents a function such as, for instance, f1, f2,
f3, etc., in Fig. 1(a). 3.3. Associativity functional networks
 A set of directed links. These are the links which connect
various input, output, and intermediate units in accor- By using the basic theory of functional equations, any multi-
dance with the advance of the FN algorithm. The arrow of a input network with s inputs (x1, x2,/, xs) and one output (xsþ1)
link specifies the direction of information flow. can be transformed to an equivalent simplified network called,
associativity functional network (Castillo and Ruiz-Cobo, 1992;
Castillo et al., 2000b). For instance, Fig. 1(b) presents the
3.2. Working with functional networks (FNs) technique associativity functional network of a typical functional
network shown in Fig. 1(a). The general expression for an
The functional networks (FNs) technique executes the associative functional network built for an independent
following steps to build a predictive model from the data at variable v is as follow
hand.
X
m
fv ðxv Þ ¼ avi fvi ðxv Þ (1)
 Step-1 (statement of the problem). This step involves the i¼1

understanding of the problem to be solved.


where avi is model coefficient, fvi is the shape function of
 Step-2 (selection of the initial topology). Unlike the artificial
variable v (it can be any admissible function like polynomial,
neural networks (ANNs) procedure, where the topology is
exponential, trigonometric, etc.), and m is the degree of
selected by a repetitive trial and error method, the initial
functions used.
topology of the functional networks (FNs) is selected on the
With increasing degrees (m) of the shape function, the
basis of properties of the problem (problem-driven design).
performance of a FN model improvises further, and the cor-
This process usually leads to the development of a single
relation coefficient (R) between actual and predicted outputs
and clear network structure.
often attains a value of as high as 1.0. This, on the other hand,
 Step-3 (simplification of the initial network). The initial
leads to the formulation of a very much complex equation
network, attained in step-2, is simplified using the func-
(model). Thus, a trade-off between model complexity and its
tional equations (Castillo and Ruiz-Cobo, 1992). FN
prediction precision (in terms of R value) is made to determine
algorithm assesses if there exists a more simplified
the suitable value of m. This is one of the common demerits of
functional network which is capable of providing the
using functional network technique. Using the input functions
same output for every given input. Such simpler networks
shown in Eq. (1), the output function of Fig. 1(b) shown above,
are called as equivalent FNs of the complex networks.
for instance, can be obtained as g(x0) ¼ g1(x1) þ g2(x2) þ g3(x3).
 Step-4 (uniqueness of representation). An investigation is
Hence, the generalized equation for finding the output
carried out to ensure if the equivalent functional network
function from an associativity functional network with s
is a unique representation of the problem. For FN topology,
inputs (x1, x2,/, xs) and one output (xsþ1) is as follow (f is
there often exist several sets of neurons (functions) those
written instead of g)
lead to exactly same output for any set of inputs. In order to
avoid estimation problems, what conditions hold for the fsþ1 ðxsþ1 Þ ¼ f1 ðx1 Þ þ f2 ðx2 Þ þ / þ fs ðxs Þ (2)
uniqueness must be decided beforehand.
where f($) represents different functions in the storing units.
 Step-5. This step involves the collection of data sets for
Consequently, the error in the jth data (ej) can be repre-
learning of the network.
sented as follow
 Step-6 (parametric learning). The neuron functions are
estimated (learned) using the collected data sets and ej ¼ f1 ðx1 Þ þ f2 ðx2 Þ þ / þ fs ðxs Þ  fsþ1 ðxsþ1 Þ (3)
combination of given functional families. The learning
The sum of squared errors En, for all n data sets can then be
may be linear or nonlinear based on the linearity of the
represented as follow
neuron functions obtained.
 Step-7 (model validation and use). In this step, the test for X
n

prediction precision and/or cross-validation of the model En ¼ e2j


j¼1
is performed. To see whether or not the selected family of ( )2
approximating functions is adequate, the checking of X
n X
m
        
¼ ai fi x1j þ fi x2j þ / þ fi xsj  fi xsþ1;j (4)
the prediction error is important. A cross-validation of j¼1 i¼1
the model is also convenient in this regard. If the
508 J. Traffic Transp. Eng. (Engl. Ed.) 2017; 4 (6): 503e521

To estimate the FN model coefficients, En can be minimized  Geometric design. The geometric designs of roadways
and subjected to include number of lanes, lane width, shoulder width, and
carriageway width, etc. The change in geometry of a
X
mk
fk ðx0 Þ ¼ aki fki ðx0 Þ ¼ ak k ¼ 1; 2; /; s þ 1 (5) roadway leads to a considerable change in its performance.
i¼1 Hence, a wide dissimilarity in the geometric details of the
selected road segments was very much important for
where x0 and ak are constants.
developing a well generalized BLOS model.
By using the Lagrangian multipliers, an auxiliary function
 Traffic volume and composition. Traffic volume and its
can be defined as follow
composition have a considerable influence on the riding
" #2 " #
X
n X
sþ1 X
m
  X
sþ1 X
m quality of bicyclists. These measures normally change with
El ¼ aki fki xkj þ lk aki fki ðx0 Þ  ak (6) differing transportation facilities in different states of a
j¼1 k¼1 i¼1 k¼1 i¼1
nation. Thus urban road segments from different states of
The minimum value of El can be obtained by using Eq. (7). India were included in the study corridors.
" #  Land-use pattern. The roadside land-use pattern has a
vEl Xn sþ1 X
X m
    considerable effect on travel demand and several other
¼2 aki fki xkj fkr xjk þ lk fkr ðx0 Þ ¼ 0
vakr j¼1 k¼1 i¼1 (7) factors. This, on the other hand, greatly affects the quality
k ¼ 1; 2; /; s þ 1; r ¼ 1; 2; /; m of bicycling environments. Thus, varieties of roadside de-
velopments such as residential area, rural fields, office or
institutional area, commercial area, and industrial area
vEl X m
¼ aki fki ðx0 Þ  ak ¼ 0 k ¼ 1; 2; /; s þ 1 (8) were included in the study corridors.
vlk i¼1

The systems of linear equations shown in Eqs. (7) and (8) In this study, required data sets were collected from four
have (km þ k) numbers of equations and (km þ k) numbers of different mid-sized cities of India namely: Bhubaneswar,
unknowns, which can be solved to get the FN model co- Rourkela, Rajahmundry, and Kottayam. These cities are
efficients, aki, where k ¼ 1, 2, /, s þ 1 and i ¼ 1, 2, /, m. Further located in eastern and southern parts of India (as shown in
details on FN procedure along with some example Fig. 2).
problems can be found in Castillo (1998), Castillo and Ruiz- Bhubaneswar is the capital and largest city of the Indian
Cobo (1992), and Castillo et al. (1998, 2000a, 2000b). In this state of Odisha. It is primarily an administrative city, and
study, the FN technique was implemented using MATLAB center of economic and religious importance in the eastern
(MathWorks, Inc., 2016) for the study analysis and BLOS part of the country. Being one fastest-developing city, it has
model development. emerged as an information technology (IT) and education hub.
Roadways in the city are generally wide and have the provi-
sion of 2e3 traffic lanes in each direction of travel. The road
networks often have the provision of separated sidewalk,
4. Site selection and data collection median and parking lane facilities. Roads passing through
interior regions of the city are generally 2-lane undivided
In order to develop a well generalized BLOS model, it was roads where the usual traffic volume is noticeably lower.
highly essential to have a precise and sufficient data set Alternatively, Rourkela city is another large city in the same
collected from various categories of road environments. Thus, Indian state. The city is popularly known as the “steel city” of
a large number of road segments (i.e., 74) were included in the Odisha because the Rourkela Steel Plant (RSP) is the first ever
study corridors, and the collected data sets were utilized for steel plant integrated into the public sector in India. The road
necessary reconnaissance. The major principle laid for infrastructure of the city is often known for a well-designed
selecting these sites are as follows. ring road (6.5e11 m wide) joining all the places of economic
importance. Roadways passing through market areas and
 Location of sites inside the city. The location of a road in- other interior areas of the city are mostly 2-lane undivided
side the city has a huge influence on the traffic volume, roads. These roads often look like an open market where on-
traffic composition and several other properties of the street parking turn-over is normally very high.
roadway. Considering a roadway situated at residential From the remaining cities, Rajahmundry is one of the
area, the proportion of independent mode of traffic would major cities in the Indian state of Andhra Pradesh, well known
be predominant, and the traffic volume would be quite less. for its historic agricultural, economic and cultural back-
Conversely, a commercial area would attract a significant grounds. It is popularly called as the “Cultural Capital” of the
number of trips from different parts of the city and would state. Kottayam is a municipal town in the Indian state of
result in higher traffic volume on the nearby roadways. Kerala and is the administrative capital city of Kottayam dis-
Similarly, the movement of heavy vehicles is often trict. In these two cities, 2- and 4-lane roads contribute the
restricted inside the city. On the other hand, the flow of most to the road networks and 6-lane roads are few in
heavy vehicles connecting from one place to another or for numbers. The road infrastructures often have the provision of
industrial purpose is significantly expected at the outskirts the sidewalk, shoulder, and median facilities. Fig. 3 shows
of a city. Hence road segments from different parts of the pictures taken from some typical road segments under
cities were included in the study corridors. varying conditions.
J. Traffic Transp. Eng. (Engl. Ed.) 2017; 4 (6): 503e521 509

Fig. 2 e Locations of studied cities in India.


510 J. Traffic Transp. Eng. (Engl. Ed.) 2017; 4 (6): 503e521

Fig. 3 e Typical study corridors representing wide variations in the bicycling environments. (a) Bidyut marg, Bhubaneswar. (b)
Ramalayam junction to NH 16, Rajahmundry. (c) Church gate e municipal stadium, Rajahmundry. (d) Kachery road, Rourkela.

4.1. Collection and extraction of input variables unauthorized stoppages of intermittent public transits were
rated using a 3-point scale where, 1 means high, 0.5 means
BLOS model development in the present context was chal- medium and 0 means minimal.
lenged with the availability of road data such as geometric Due to the heterogeneous traffic flow on Indian roads, the
details, traffic volume, and traffic speed, etc. The historical average operating speed is significantly lower than the same
data of urban roads are mostly unavailable with the city au- in developed countries under homogeneous traffic conditions.
thorities in mid-sized cities of a developing country like India. In addition to this, a large variation exists among the speeds of
Though a few data sets are available, those are poorly individual vehicles under heterogeneous traffic conditions.
maintained and probably not useful for the research work. As Thus, the speed measures such as spot speed or space mean
a consequence, the field investigation procedure was adopted speed that are normally calculated for homogeneous traffic,
to gather the required data set. Instruments such as should not be considered for heterogeneous traffic situations.
measuring tapes, high-definition (HD) video cameras, and In view of this, the mid-segment traffic flow was videotaped
tripod stands were used during the inventory survey to carry over an elongated longitudinal trap of 30 m. Expected peak 2 h
out several geometrical measurements and videotaping of of traffic flow (AM 8:30e10:30 or PM 4:30e6:30) were selected
the traffic flow. Widths of the outermost lane, shoulder, for the data collection as bicyclists encounter the worst con-
paving between the outermost lane stripe and the edge of ditions during this period. Recorded videos were projected on
pavement, median, curb and gutter were measured during a large screen in the laboratory and the time taken by the
the inventory survey. Other information such as: the pres- usual motorists to cross this trap was extracted with an ac-
ence of proper lane markings, curb, sidewalk, shared-use curacy of 0.1 s. Using this data average traffic speed (ratio of
path in usable conditions, and trees on roadsides were average time taken by motorists to cross the trap and length of
collected using a 2-point scale where 1 means present and the trap) on each segment was calculated.
0 means not present. Number of driveways merged with the Team members also extracted several other desired data
segment were counted manually by traveling from one end of sets from these videos such as: peak hour volume (PCU/h),
the segment to another; and how many of those driveways pedestrian volume (ped/h), percentage of heavy vehicles (%)
are carrying a high volume of traffic was also inspected. and approximate vehicular ingresseegress to the on-street
Pavement surface quality of each segment was rated parking area (veh/h/km). For the estimation of peak hour
using a 5-point scale where 5 means excellent and 1 means volume (PHV) from 2 h of volume counts, running average
worst. Roadside commercial activities and interruptions by method was used. Equivalent passenger car units (PCUs) of
J. Traffic Transp. Eng. (Engl. Ed.) 2017; 4 (6): 503e521 511

vehicles endorsed in Indian Road Congress (IRC, 1990) were explanations on these two survey approaches are presented in
used, and traffic volume was expressed in PCU per hour. the following subsections.
Some notifiable diversity in the extracted road attributes are
3e14 m wide carriageway with 1e4 traffic lanes in each 4.2.1. Videography survey
direction of travel, 250e5100 PCU/h traffic volume during In order to obtain quite precise responses, two eligibility criteria
peak commuting hours, 23e46 km/h of operating speed, and were used for the users willing to participate in this survey such
0e7% heavy vehicles in the total traffic flow, etc. The as: (1) a participant should be at least 14 years old, (2) a partic-
detailed statistics of important road attributes including ipant should have at least a year of cycling experience on city
their range, mean, and standard deviation are highlighted in roads. Roughly 150 participants were recruited from the
Section 5 (Table 2). campus and nearby localities of an institution of national
importance through a common e-mail explaining the survey
4.2. Perception survey and collection of output variable objectives and its importance. The participants were asked to
fill up an online registration form and show their interest in this
An extensive perception survey was carried out in this study to survey. The registration form was designed so as to collect
assess the perceived satisfactions of on-street bicyclists under valuable information on the age, sex, geographic origins,
prevailing road conditions. Literature revealed that mainly two working class, and social, economic and educational back-
types of methods are being used for this purpose such as: grounds, etc., of the possible participants. Table 1 summarizes
traveler intercept survey (on-site face to face interaction with the wide diversities observed in the participants' socio-
the road users of interest) and videography survey (showing demographic details and bicycling characteristics (bicycling
roadway environments to the participants through suitable experiences, daily average bicycling distance, bicycle trip
video clips and asking them to rate the road under shown purpose and types of users).
conditions). The former approach is often preferred to obtain In this study, the videography survey was conducted in
more realistic responses from the participants who have three different sessions, and around 50 participants partici-
immediately traveled on the roadway under investigation. pated in each session. One typical video clip having 30e45 s of
However, this method is very much time consuming and loss- duration was selected for each segment to exemplify the usual
making particularly while gathering large sets of data. Thus, the traffic volume, traffic speed, congestion and stopping interval
videography survey approach is often preferred as an alterna- of public transits, etc., on that segment. A brief overview on
tive approach in various research works. For instance, bicycle surrounding environmental characteristics (roadside de-
compatibility index (BCI) model (Harkey et al., 1998) and bicycle velopments and on-street parking activities, etc.) was also
model (Jensen, 2007) are two of the successful models built upon featured in the clips. Each clip was shown on a large screen and
the videography survey data. Harkey et al. (1998) tried to participants were asked to rate the segment for bicycle use
validate a video-based methodology using a stationary under shown conditions. The question asked to the partici-
camera, and concluded that the video-based methodology is a pants to assess their satisfaction levels was kept as simple as:
valid technique for obtaining realistic perspectives of how much satisfied are you as a bicyclist on the roadway
bicyclists. The video-based methodology also has following shown?” A 6-point Likert scale (1 meas extremely satisfied and
several advantages over traveler intercept survey (Jensen, 2007). 6 meas extremely dissatisfied) was used to collect these
ratings.
 The number of street segments that participants can rate During the videography survey, some “repeater clips” were
during a reasonable period of time is generally high. used to assess the individual respondent's ability to detect
 More diverse group of participants can be included. minor changes and to deliver identical responses. Repeater
 It is more cost-effective than having respondents on site. clips were videoed at the same part of the segment as its
 The exact same roadway and traffic conditions, for original, but with differing traffic volumes. These clips were
example, may be experienced by many respondents; the shown to the survey participants at a fixed interval. During the
conditions to be rated can be chosen from several video- data analyses sections, it was observed that around 10 persons
tapes of the same roadway segment. This form of variable were not able to detect the minor changes in repeaters and had
control is impossible when respondents actually walk and provided inappropriate responses. Thus, any information ob-
ride on the roadway. tained from these 10 participants were not included in the
 There are no traffic risks to the respondents, which makes model building process. As such, this survey resulted in a total
it easier to include roadway segments that may include of 10,360 effective ratings (i.e., 74 segments  140 effective
high perceived risks. participants). These ratings were termed as “perceived BLOS
scores”. An investigation was carried out to check whether
In light of the above advantages, the present study has 10,360 numbers of perceived BLOS scores are sufficient for
implemented the videography survey approach to accumulate BLOS model building. The allowed error in the estimation of
large quantity of data sets required in the BLOS model build- the mean BLOS score (3.49) was calculated by using the
ing. In order to obtain reliable responses in this survey, Cochran's sample size formula (Cochran, 1977). At 95%
enough care was taken which are discussed in the following confidence level, the error value was observed to be 0.22%.
section in detail. Besides, a traveler intercept survey was also This means that with a total of 10,360 responses the
carried out at some identified locations (i.e., 20 diversified percentage of error in the mean score is 0.06 (0.22/3.49) only.
segments randomly selected from the study area) to compare This minimal error well implied the sufficiency in the
the results with videography survey-based data sets. Detailed collected data sets.
512 J. Traffic Transp. Eng. (Engl. Ed.) 2017; 4 (6): 503e521

Table 1 e Participants' characteristics in the survey sample.


Attribute Distribution Frequency Percentage (%)
Gender Female 66 44.0
Male 84 56.0
Age (years old) 20 7 4.6
21e30 57 38.0
31e40 39 26.0
41e50 26 17.4
51e60 15 10.0
60 6 4.0
Education Matric or less 28 18.7
Intermediate 48 32.0
Graduate 62 41.3
PG or above 12 8.0
Experience with bicycling (year) <5 52 34.7
5e10 63 42.0
>10 35 23.3
Daily average bicycling distance (km/d) <5 43 28.7
5e10 83 55.3
11e20 21 14.0
>20 3 2.0
Trip purpose Recreation 41 27.3
Exercise or health 34 22.7
Personal errands 27 18.0
Education 24 16.0
Commuting to/from work 12 8.0
Visit a friend or relative 9 6.0
Other 3 2.0
User type Regular rider 64 42.6
Occasional rider 86 57.3
Total for each attribute 150 100

4.2.2. Traveler intercept survey network (FN) technique, (4) goodness-of-fit and prediction
As explained earlier, the traveler intercept survey was carried precision of the developed model evaluated through the
out on 20 roadway segments located in different parts of the application of several statistical parameters, (5) comparison of
study area and representing differing road environments the newly developed model with other existing models, and
(excellenteworst). The survey principles were kept as similar (6) defining the ranges of service categories (AeF).
as to that of the videography survey in terms of number of
responses, diversities in participants' socio-demographics and 5.1. Checking the reliability of videography survey data
other characteristics. This survey was conducted during the
peak hours of traffic flow as the data sets to be obtained were The videography survey data obtained from 20 road segments
desired to reflect the worst conditions observed on the road- were used to check the reliability of the videography survey
ways. At least 140 bicyclists from each segment were inter- method. The overall BLOS score perceived by survey partici-
viewed by the team members. The percentage distributions in pants on a particular road segment was calculated by taking
participants' gender, age, education level, experience with the average of all perceived BLOS scores (effective scores only)
bicycling, daily average bicycling distance, trip purpose, and obtained for that segment. Here, overall perceived BLOS
user type (i.e., regular rider or occasional rider) were well scores for each segment as per traveler intercept survey data
diversified and as similar as those shown in Table 1. Each and videography survey data were estimated separately.
participant was asked to rate the roadway segment he/she These two sets of data were compared with each other to
immediately traveled on. The question asked to obtain the witness the level of similarity that does exist in between them.
participant's response was kept as simple as “what is your The correlation coefficient (R) between these overall scores
perceived level of satisfaction while riding on the road obtained for 20 segments was as high as 0.96, and average
segment?” The same 6-point Likert scale was used, and 140 absolute error (AAE) was as less as 0.17. Hence, it was ensured
effective perceived BLOS scores were collected from each that the videography survey data collection method for
segment. operational analysis of bicycle facilities could be relied like
field intercept survey method.

5. Analyses, results, and discussions 5.2. Variables selection

This section gives a detailed discussion on: (1) checking reli- An extensive Pearson's correlation analysis was carried out to
ability of videography survey data, (2) selection of influencing identify which road attributes are significantly (p < 0.001)
variables, (3) development of the BLOS model using functional affecting the perceived satisfactions of on-street bicyclists.
J. Traffic Transp. Eng. (Engl. Ed.) 2017; 4 (6): 503e521 513

The wide-range of road attributes collected in this study were

Standard deviation

Note: Average effective width of outermost through lane (WOTLE) is defined as Wt þ Wl þ Wr, where Wt is sum of outermost lane width and paved shoulder width, Wl is width of paving between
used as the independent variables, and the array of overall
perceived BLOS scores obtained for individual segments in the

1442.45
videography survey was used as the dependent variable. Sig-

489.10
0.71
0.90

4.88
0.37

0.44

0.39

0.95
nificant variables identified from the Pearson's correlation
analysis were used as the set of independent variables in the
BLOS model building process. Table 2 shows the notations and
descriptive statistics of all output and input variables used in
this study for the development of BLOS model.

Mean

1069.6

745.78
Though variability in the traffic speed was anticipated to be

33.94
3.49
3.68

3.85

0.49

0.45

0.83
an important parameter, it had a high correlation of 0.72 with
average traffic speed. Thus, in order to avoid autocorrelation
among independent variables any one parameter was to be

1.55e5.40

148e2586
Range
chosen. Average traffic speed parameter was preferred in this

2.5e4.5

0e6000
23e46
study as it was more correlated with the BLOS score. Indian

2e7

0e1

0e1

0e3
people generally do not prefer to walk for longer distances.

outermost lane stripe and edge of pavement, and Wr is average width reduction due to encroachments in the outermost lane (Landis et al., 1997).
Moreover, a significant percentage of the city roads have the
provision of sidewalk facilities those offer separate space for

1e5 scale (5 means excellent,


pedestrian use and restrict them from interfering with the
mainstream traffic flow. These are some probable reasons due

Unit/scale
to which the influence of “pedestrian volume” parameter was

6-point Likert scale


found to be insignificant. Similarly, the frequency of drive-

1 means worst)

Number(s)/km
0.5 (medium),

0.5 (medium),
ways also had an insignificant correlation with the perceived

0 (minimal)

0 (minimal)
PCU/h/lane

veh/h/km
BLOS score (p > 0.001). However, the frequency of driveways

1 (high),

1 (high),
carrying a high volume of traffic had a significant correlation

Meter

km/h
(p < 0.001) with the perceived BLOS score. This means bi-
Table 2 e Notations and descriptive statistics of variables used in the BLOS model development.

cyclists in the present context are perceiving considerable


dissatisfaction under the influence of driveways carrying a
Abbreviation
BLOSPerceived

high volume of traffic, and are okay with those carry lower
volume of traffic. Thus, the former parameter was retained in
WOTLE
PHVL

the model development process.

IIPT
PCI

The correlation between individuals' perceived BLOS CA

D
S

P
scores and socio-demographic variables was not significant
with p  0.001. However, there were some tendencies. Men
Output
Type

seemed to be more satisfied than women. Elderly people


Input
Input
Input
Input

Input

Input

Input
Input
seemed to be more dissatisfied than youth. However, quanti-
tative variables (geometric, traffic and built environmental
variables) were observed to be dominating over socio-eco-
Vehicular ingresseegress volume to on-street parking area

nomic attributes and driver characteristics (e.g., gender,


Frequency of driveways carrying a high volume of traffic

experience level, household size, income, working class, etc.)


Variable (subjected to specific direction)

Average effective width of outermost through lane

in the present context. This happened probably because, un-


like in high-income cities the discrepancies in living stan-
dards, income, and other socio-demographics of the people in
Interruptions by unauthorized stoppages of

mid-sized cities of a developing country like India is not very


Commercial activities on roadside area

inclusive. Similar observations are also documented in several


studies carried out in developed countries such as USA (Landis
Peak hour traffic volume per lane

et al., 1997; TRB, 2000, 2010; etc.), and Denmark (Jensen, 2007),
intermittent public transits

etc. Thus, quantitative variables are only considered in this


Pavement condition index

study for the development of BLOS model.


Perceived BLOS score

Average traffic speed

5.3. FN based BLOS model development and sensitivity


analysis

In this study, the FN technique was implemented using


MATLAB (MathWorks, Inc., 2016) for the development of BLOS
model. Data of randomly selected 52 segments (70% of total)
Sl. No.

were used for the model training, and remaining data were
used for the model testing. By using the following Eq. (9), all
1
2
3
4
5

8
9

dependent and independent variables were normalized


514 J. Traffic Transp. Eng. (Engl. Ed.) 2017; 4 (6): 503e521

within the range of [0, 1] to avoid their dimensional effects in


the model building process.

v  vmin
vNormalized ¼ (9)
vmax  vmin
where v is any variable, vNormalized is normalized value of v,
vmax is maximum value of v, vmin is minimum value of v. The
maximum and minimum values of each variable used in this
study are shown in Table 2.
In this study, the functional networks (FNs) based BLOS
model were developed by selecting an appropriate shape
function (SF) (f) and its proper degree. Performances of five
different SFs such as: polynomial, exponential, sin($), cos($) and
tan($) were investigated, and the obtained results are shown in
Fig. 5 e Associativity functional network for the proposed
Fig. 4. This figure depicts that, the polynomial function is
BLOS model.
producing the highest correlation coefficient R value (between
predicted and perceived BLOS scores) at all lower degrees of
SFs (1e5). Though all shape functions are providing highly
precise results with Ry1:0 at degrees of 10 and higher, a WOTLE, for instance, are 1.1717, 2.0221, 0.0592 and
lower degree is always desired as it leads to the formation of 1.2315 at degrees of 1e4, respectively. Thus, the associative
a less complex model. Complex structured models require functional network for the variable (WOTLE) would be e
higher computational efforts for the field implementations 1.1717WOTLE þ 2.0221WOTL2E  0.0592WOTL3E  1.2315WOTL4E.
and thus are undesired. Thus, a polynomial SF was chosen The associative functional network for remaining seven
for the development of a less complex BLOS model. After variables can also be built in a very similar way.
choosing the suitable shape function, the next step was to An empirical equation shown below is built using Eqs. (1)
decide its proper degree. In order to accomplish this, a trade- and (2) to predict the normalized values of BLOS scores
off was made with the help of Fig. 4 between the degree and (BLOSNorm).
complexity of the BLOS model, and a polynomial SF with " # " #
X
s X
m
  X
s X
m
 
degree “four” was adopted. As observed in Fig. 4, the BLOSNorm ¼ aji fji xj ¼ a0 þ fi xj (10)
polynomial SF is providing a R value of approximately 0.94 j¼1 i¼1 j¼1 i¼1

which is a very high and well-acceptable value.


where s is number of independent variables (s ¼ 8), m is degree
Using the concepts of associativity functional networks, the
of SF (m ¼ 4).
network shown in Fig. 5 was built to predict the overall BLOS
Using the model coefficients shown in Table 3, above Eq.
scores of urban street segments carrying heterogeneous
(10) is expanded as follows to find the normalized value of
traffic. In Fig. 5, f1 ¼ f2 ¼ / ¼ f8, f1 e f8 are polynomial
predicted BLOS score for any urban street segment.
functions with degree “four”. The FN algorithm was executed
in MATLAB to estimate the model coefficients and the a0 ¼ 0:504 (11)
constant term (a0), and the obtained results are presented in
Table 3. Columns 2e8 of Table 3 contain the coefficients of X
m

individual independent variables at degrees of 1e4. An fi ðx1 Þ ¼ 1:1717WOLTE þ 2:0221WOTL2E  0:0592WOTL3E


i¼1
associative functional network for an input variable (v) can be
 1:2315WOTL4E
built by putting the corresponding coefficient estimates in Eq.
(1). As observed in Table 3, the coefficients of a variable (12)

X
m
fi ðx2 Þ ¼ 0:3067PHVL þ 0:8544PHVL2  2:4434PHVL3
i¼1

þ 1:5336PHVL4 (13)

X
m
fi ðx3 Þ ¼ 0:0218S  0:4736S2 þ 1:4273S3  0:9338S4 (14)
i¼1

X
m
fi ðx4 Þ ¼ 1:0866PCI þ 6:2554PCI2  11:0284PCI3 þ 5:7909PCI4
i¼1

(15)

X
m
fi ðx5 Þ ¼ 0:0383CA þ 0:1412CA4 (16)
Fig. 4 e Performance of shape functions (SFs) of the
i¼1
functional networks (FNs) approach.
J. Traffic Transp. Eng. (Engl. Ed.) 2017; 4 (6): 503e521 515

Table 3 e Functional network based BLOS model coefficients.


Degree Variable Constant (a0)
WOTLE PHVL S PCI CA IIPT P D
1 1.1717 0.3067 0.0218 1.0866 0.0383 0.1045 1.1174 0.0152 0.5040
2 2.0221 0.8544 0.4736 6.2554 0.0000 0.0000 4.9067 0.1258
3 0.0592 2.4434 1.4273 11.0284 0.0000 0.0000 8.2134 0.0000
4 1.2315 1.5336 0.9338 5.7909 0.1412 0.0684 4.2785 0.0478

fmin(xi), Si is sensitivity in percentage, and n is the number of


X
m input variables. fmax(xi) and fmin(xi) values were calculated by
fi ðx6 Þ ¼ 0:1045IIPT  0:0684IIPT4 (17) feeding the developed model with maximum and/or mini-
i¼1
mum values of ith variable and mean values of remaining
variables.
X
m
fi ðx7 Þ ¼ 1:1174P  4:9067P2 þ 8:2134P3  4:2785P4 (18) It is obvious that, the higher the value of Si greater is the
i¼1 importance of corresponding input variable in predicting the
output variable. Thus, the variable WOTLE having the highest
X
m
fi ðx8 Þ ¼ 0:0152D þ 0:1258D2  0:0478D4 (19) value of Si (38.3%) has been assigned with a rank of “1” in
i¼1 Table 4. Remaining variables are also ranked accordingly to
show their relative importance in BLOS score prediction in
In Eqs. (12)-(19), values of input variables are their
the present context. WOTLE has the highest importance
normalized values between 0 and 1. The sum of outputs from
possibly because, outermost lane and paved shoulder are the
Eqs. (11)-(19) gives the normalized value of the predicted BLOS
spaces where bicyclists actually move. On Indian roads,
score for the desired street segment. Finally, the de-normal-
where left-hand drive rule is followed, bicycle users usually
ized value of the predicted BLOS score (BLOSPredicted) of the
prefer to move on the outermost lane. Thus, the perceived
segment can be found out by using the following Eq. (20).
satisfaction levels of the bicycle users are largely influenced
BLOSPredicted ¼ BLOSNorm ðBLOSMax  BLOSMin Þ þ BLOSMin (20) by the available effective width of outermost lane. As
observed in Table 4, peak hour traffic volume per lane (PHVL)
where BLOSMax is maximum value of perceived BLOS scores
has the second highest influence on bicycle service quality.
(BLOSMax ¼ 5.40), BLOSMin is minimum value of perceived
This may be because several categories of vehicles widely
BLOS scores (BLOSMin ¼ 1.55).
varying in their size and operating conditions share the
The collection of the BLOS model inputs from real field
entire roadway width with a lack of lane discipline. Thus, on-
conditions is not very complicated. Thus, the developed
street bicyclists encounter a very complexly interaction with
model has a good scope for its easy implementations in the
the vehicular traffic flow. Thus, PHVL parameter hugely
real practice. The city planners and engineers can use this
influences the quality of the bicycling environment in
model for various purposes including the assessment of urban
developing countries. Similarly, the vehicular ingresseegress
on-street bicycling environment and the bicycle route plan-
to the on-street parking area obstruct the path of bicyclists
ning in mid-sized cities. The sensitivity analysis was essential
and significantly degrade their riding quality. This variable
as it reports the relative importance of input variables based
was observed to be the third most important variable in the
on their percentage contribution to the prediction of the BLOS
present context. Roadside commercial activities (CA),
score. For this purpose, a sensitivity percentage value (Si) was
pavement condition index (PCI), frequency of driveways
estimated for each modeled road attribute using Eqs. (21) and
carrying a high volume of traffic (D), average traffic speed (S),
(22) (Gandomi et al., 2013) and results are summarized in
and unauthorized stoppages of intermittent public transits
Table 4.
(IIPT) are the remaining significant variables stated in a
Ni ¼ fmax ðxi Þ  fmin ðxi Þ (21) descending order of importance.

Ni 5.4. Goodness-of-fit and prediction precision of the


Si ¼  100% (22)
P
n
developed model
Ni
i¼1

where fmax(xi) is the maximum value of the predicted output An extensive investigation has been carried out in this study
over ith input, fmin(xi) is the minimum value of predicted to assess the goodness-of-fit of the developed model with
output over ith input, Ni is the difference between fmax(xi) and training data sets and its prediction precision with testing
data sets through the application of various statistical pa-
rameters. Fig. 6 shows a plot of FN model predicted BLOS
Table 4 e Relative importance of input variables. scores vs. overall perceived BLOS scores (both for training
data and testing data sets). The figure depicts that, the
Variable WOTLE PHVL S PCI CA IIPT P D
scatter of predicted scores is well within the 80% prediction
Si (%) 38.3 21.8 3.7 5.9 8.9 3.2 12.7 5.5 limit. The model also justifies its high prediction precision in
Rank 1 2 7 5 4 8 3 6
the present context with a high correlation coefficient (R)
516 J. Traffic Transp. Eng. (Engl. Ed.) 2017; 4 (6): 503e521

(ii) Mean (m) and standard deviation (s) of BLOSPredicted/


BLOSPerceived (ratio of predicted and overall perceived
BLOS scores)

Mean (m) and standard deviation (s) of the ratio of predicted


and actual outputs immensely indicate the prediction preci-
sion of an empirical model. Under ideal conditions, i.e., when
model predicted outputs are exactly equal to the actual ones, m
becomes 1.0 and s becomes 0. For the developed BLOS model,
m values are 1.004 and 1.001 for training and testing data sets
respectively, which are very close to 1.0. Besides, s values are
Fig. 6 e Plot of FN model predicted BLOS scores vs. overall 0.067 and 0.085 for training and testing data sets respectively
perceived BLOS scores. which are very close to 0. Thus, it can be concluded that the
developed BLOS model has a very precise performance in the
present context.
between predicted and overall perceived BLOS scores
(R ¼ 0.940 for training data, and R ¼ 0.943 for testing data). (iii) 50% and 90% cumulative probabilities (P0 50 and P0 90) of
Thus, the developed model has a high prediction precision BLOSPredicted/BLOSPerceived
in the present context.
As the R parameter has been criticized by many researchers Value of 50% cumulative probability (P0 50) below “one” im-
as being a biased estimate, several other statistical parameters plies under prediction and above “one” implies over-predic-
were also applied to thoroughly assess the goodness-of-fit and tion by a predictive model. Similarly, 90% cumulative
prediction precision of the developed model. The parameters probability (P0 90) value reflects the variation in the ratio of
are such as (i) NasheSutcliffe efficiency coefficient (E), (ii) mean predicted and actual outputs for the total observations. Thus,
(m) and standard deviation (s) of BLOSPredicted/BLOSPerceived (ratio values of these two parameters closer to “one” indicate a high
of predicted and overall perceived BLOS scores), (iii) 50% and prediction efficiency. In order to estimate the values of these
90% cumulative probabilities (P0 50 and P0 90 respectively) of parameters, BLOSPredicted/BLOSPerceived values for all segments
BLOSPredicted/BLOSPerceived, (iv) error measuring parameters and were arranged in ascending order (separately in training data
the overfitting ratio (OR). A detailed discussion of each aspect is and testing data sets) and the cumulative probability (P0 ) was
given below. calculated by using Eq. (26).

(i) NasheSutcliffe efficiency coefficient (E) l


P0 ¼  100% (26)
sþ1
As an alternative measure to R, NasheSutcliffe efficiency where l is serial or index number of a segment.
coefficient (E) is also used for the best-fit calculations of the The variation of BLOSPredicted/BLOSPerceived with the cumu-
predicted and actual output. This parameter compares the lative probabilities is shown in Fig. 7. It can be observed that,
predicted and actual values of the output variable, and eval- the P0 values are very close to 1.0 at any value of percentage
uates how good the model is in explaining the variance in total probabilities. As estimated from Fig. 7, the values of P0 50
data sets. A NasheSutcliffe efficiency coefficient value of 1.0 parameter are 0.991 and 0.983 for training and testing data
represents an exact explanation of the variance. For the sets respectively. Similarly, the values of P0 90 parameter are
developed BLOS model, the E value was estimated using Eqs. 1.085 and 1.095 for training and testing data sets
(23)-(25) (Nash and Sutcliffe, 1970) and was found to be 0.884 respectively. As these estimated values are all close to 1.0,
for both training and testing data sets. Such a higher value
of E (closer to 1.0) signifies an excellent performance of FN
model in the present context.

E ¼ ðE1  E2 Þ=E1 (23)

s 
X 2
E1 ¼ BLOSPerceived  BLOSPerceived (24)
i¼1

X
s
E2 ¼ ðBLOSPerceived  BLOSPredicted Þ2 (25)
i¼1

where BLOSPerceived, BLOSPerceived , BLOSPredicted are perceived,


average of perceived and predicted values of BLOS scores,
respectively, s is total number of road segments (i.e., 52 in Fig. 7 e Cumulative probability plot of BLOSPredicted/
model training and 22 in model testing). BLOSPerceived (ratio of predicted and overall perceived BLOS
scores).
J. Traffic Transp. Eng. (Engl. Ed.) 2017; 4 (6): 503e521 517

the developed model has a very precise prediction

Overall
performance at both 50% and 90% cumulative probabilities.

rank

2
3
4
(iv) Error measuring parameters and overfitting ratio (OR)

RMSE
0.219
0.264
0.603
1.535
1.998
The use of various error measuring parameters such as
average absolute error (AAE), maximum absolute error (MAE)
measuring
parameter
and root mean square error (RMSE) is very common in any field
Error

MAE
0.549
0.664
2.062
2.759
3.643
of science and engineering. Values of these parameters as
estimated using predicted and overall perceived BLOS scores
0.176 are presented in Table 4. As observed, the numerical values of
0.210
0.487
1.433
1.775
AAE

these parameters are significantly lower and thus well


satisfying to the criteria of a reliable model. Quantitatively,
the generalization of any empirical model is presented as the
probability of BLOSPredicted/BLOSPerceived

overfitting ratio (the ratio of RMSE in model testing to that in


model training). The developed BLOS model has reported an
1.085
1.095
1.260
1.720
0.915
P0 90

overfitting ratio value of 1.205 which is very much close to 1.


This means that the model is well generalized and capable of
providing reliable results with new data sets.
Cumulative

It is well convinced from the above observations that, the


BLOS model developed in this study using functional network
(FN) technique is highly reliable for its applications in the
present context. The model will show an excellent perfor-
mance in a mid-sized city of any developing country provided
0.991
0.983
1.071
1.437
0.519
P0 50

that, the field observations are within the ranges of variables


shown in Table 2. In this study, the model inputs are collected
from a large number of road segments (i.e., 74), which well
represent the possible variabilities and complexities
persisting in roadway environments in mid-sized cities.
Thus, the developed model is well anticipated to perform
calculation of BLOSPredicted/BLOSPerceived

well in mid-cities of India as well as other developing


countries around the world. The present study has taken a
0.067
0.092
0.180
0.251
0.379
Table 5 e Prediction precision of BLOS models under mixed traffic conditions.

country like India as the case study which is the second


largest populous country in the world sharing about 18% of
the total population. Also, developing countries contribute to
Arithmetic

about 80% of the world's population. Hence, the proposed


model has tremendous potential for its wide applications.
However, when the city size is very large like metropolitan
area or very small like a town, it is expected that some
additional variables or lesser number of variables may
1.004
1.001
1.070
1.456
0.453
m

contribute to the bicycle service quality. In such cases, the


study methodology could be easily duplicated or calibrated
to meet the altered requirements.

5.5. Developed model vs. existing models


Best fit calculation

3.609
6.816
0.884
0.884
0.288
E

As indicated in the earlier discussion, all existing bicycle


models are built upon the data sets collected from homoge-
neous traffic situations prevailing in developed countries.
Thus, it is obvious that these model will not provide precise
0.940
0.943
0.394
0.409
0.541

results in the present context, i.e., under heterogeneous traffic


R

flow situations. A detailed investigation has been carried out


in this study to ensure about the same. Three existing and
well-followed BLOS models namely, HCM model (TRB, 2010),
Training
Testing
Data

FDOT model (FDOT, 2009) and BCI model (Harkey et al., 1998)
All
All
All

were employed to predict the BLOS scores of investigated


segments. The predicted values were compared with the
Model

actual ones, and all statistical parameters discussed earlier


FDOT
HCM

were estimated. The detailed summary of all these


BCI
FN

parametric values is presented in Table 5. It can be well


518 J. Traffic Transp. Eng. (Engl. Ed.) 2017; 4 (6): 503e521

convinced from Table 5 that, existing models are not able to road segment is offering excellent service quality (BLOS A) or
provide good results under heterogeneous traffic situations. very poor service quality (BLOS F). From study corridors,
Some of the key reasons for this happening are discussed around 4%, 46%, 43% and 7% segments are offering service
below. qualities of B, C, D and E, respectively. This indicates that the
The traffic flow situation in Indian cities is highly hetero- majority (around 95%) of studied segments in Indian mid-
geneous where the user behavior is significantly different sized cities are offering average or inferior services to bi-
from that under homogeneous traffic situations. Thus, the cyclists. The probable reasons behind this happening are
choice of significant variables for the BLOS model develop- discussed below.
ment under these differing situations is significantly different. As observed from the sensitivity analysis, WOTLE followed
The newly developed FN based BLOS model has considered by PHVL has the highest impact on the BLOS of urban street
two new variables namely, interruptions by unauthorized segments. WOTLE is primarily determined by the widths of
stoppages of intermittent public transits (IIPT) and frequency bicycle lane, paved shoulder, and the outermost lane. How-
of driveways carrying a high volume of traffic (D). In Indian ever, the study corridors identified from mid-sized cities of a
cities, the intermittent public transits (e.g., 3-wheeler autos developing country like India do not have the provision of the
and city buses) frequently stop on the roadsides for the separate bicycle lane or wide outside lane facilities. Although
boarding and drop down of people and goods. These few road segments (e.g., Nandan Kanan Road corridor of
haphazard activities were observed to hinder the path of on- Bhubaneswar City) have the provision of the shared-use path;
street bicyclists and cause significant (p < 0.001) discomfort to those are very poorly maintained, often occupied by street
them. Hence, a new parameter, IIPT, was included in the vendors, parked by motor vehicles illegally, or often poor
model building process to address these negative influences. drainage water occupies the space. In this situation, bicyclists
Similarly, the presence of driveways also creates considerable prefer to ride on the main carriageway rather than the shared-
interruptions to the movements of bicyclists. Bicyclists use path. Hence, these segments are able to offer BLOS B or
perceived to develop significant (p < 0.001) dissatisfaction and inferior although they are designed to offer better services.
frustration under the influence of driveways carrying a high The provision of wide outside lanes (or paved shoulders) at the
volume of traffic. However, while all driveways were consid- present scenario is also very rare. Moreover, the city roads are
ered together (including those carry less traffic volume), the influenced by the heterogeneous flow of 148e2586 PCU/h/
frequency of driveways parameter had an insignificant lane. Other factors such as good pavement conditions with
(p > 0.001) correlation with the BLOS score. Thus, the fre- proper lane markings, land use pattern that brings nice aes-
quency of driveways carrying a high volume of traffic only thetics to the bicycling environment and restrictions on on-
been considered in this study. These key factors have aided street parking activities are also greatly missing on these
the proposed model to have a better prediction performance study corridors. Hence, the study corridors are not able to
in the present context. offer BLOS class A. However, few road segments (around only
4%) having the provision of the 5e7 m wide outermost lane
5.6. Ranges of BLOS classes (AeF) (including paved shoulder) and carrying below 900 PCU/h/lane
traffic volume during peak hours are observably offering good
BLOS scores obtained for the study corridors were stratified services, i.e., BLOS B. These few segments are also somewhat
into six BLOS classes (A through F) using a simple concept free from on-street parking and vending activities.
commonly used in traffic engineering studies. The mean value Approximately 87% of the total studied segments are of-
of all perceived BLOS scores was approximately 3.5, which fering average and below average quality of services, i.e.,
corresponds to the boundary between BLOS classes C and D. BLOSs C and D. These segments are mostly 4-lane or 6-lane
By considering the symmetry of this boundary point, the BLOS roads and carry up to 2600 PCU/h/lane traffic volume during
scale has been determined as follows. the peak hours. Outermost lane or paved shoulder being
BLOS A (excellent service quality): BLOS score  1.5. partially occupied by street vendors, parked vehicles or
BLOS B (good service quality): 1.5 < BLOS score  2.5. roadside advertisement panels, its effective width was
BLOS C (average service quality): 2.5 < BLOS score  3.5. observed to be constricted. Pick up travelers also mostly use
BLOS D (below average service quality): 3.5 < BLOS this lane to catch intermittent public transports such as
score  4.5. shared van and 3-wheelers at their own convenience. These
BLOS E (poor service quality): 4.5 < BLOS score  5.5. all activities combined together largely degrade the riding
BLOS F (very poor service quality): BLOS score > 5.5. quality of bicyclists. Thus, necessary actions should be taken
against these trivial activities in order to achieve better BLOS
classes.
6. Model applications and observations City roads passing through the market area are mostly 2-
lane undivided and are often characterized by high on-street
The BLOS score of each studied segment was estimated parking turnover of up to 6000 veh/h/k. Kachery Road of
through the application FN model, and the corresponding Rourkela City was observed to be one of them. These road
service category (AeF) was determined by using the BLOS segments look like open markets, where vendors and cus-
scale discussed earlier. Results showed that not even a single tomers have many activities to perform on the roads. Here,
J. Traffic Transp. Eng. (Engl. Ed.) 2017; 4 (6): 503e521 519

commercial activities predominate over any other activities model developed in this study is highly reliable for its ap-
including travel. This results in the increasing pedestrian plications in mid-sized cities under heterogeneous traffic
volume, traffic congestion, on-street parking turn-over, and flow situations.
the constriction of usable road space. Here, the frequency of A sensitivity analysis has been carried out to detect the
access roads is also very high which frequently hinders the relative importance of input variables based on their per-
path of through bicyclists and increases the risk of side colli- centage contribution to the prediction of bicycle service
sions. Over-utilization of these segments also deteriorate the quality. The effective width of the outermost lane is
pavement surface quality. These conditions combined observably the most important variable in the present
together have made these few segments to offer poor level of context, and contributing 38.3% to the FNs model prediction
service, i.e., BLOS E. Road segments on which very poor service of BLOS score. Hence, the provision of a wide outer lane
quality F was expected were perhaps not observed in the and/or paved shoulder is well desired to enhance the quality
study area. Roadways connecting important places to fulfill of bicycling environment. Traffic volume on the main car-
the high functional requirements on which motorized traffic riageway is observably the second most important variable
occupies the full carriageway width and also tries to maintain and contributing 21.8% to the model prediction. Due to the
a high speed is the worst condition that a bicyclist may heterogeneous flow of motor vehicles, several small to big
encounter. Inter-city roads or origin and termination part of vehicles interact with on-road bicyclists and sufficiently
any expressway inside a city are also the locations where very degrade their perceived satisfaction levels. Thus, the mini-
poor service F can be expected. Under such scenarios, the mization of bicycle-vehicle interactions through the provi-
provision of a separate bicycle lane could provide better ser- sion of a separate bicycle lane is well desired in order to
vice levels to the bicyclists. enhance the bicycle service quality. Likewise, cross-traffic
flow from the on-street parking area obstructs the path of
through bicyclists and significantly hampers their sense of
7. Conclusions safety. This activity being the third most important variable
is contributing 12.7% to the model prediction. The remain-
The development of a reliable bicycle level of service (BLOS) ing significant variables in a descending order of importance
model being the prime objective this study, several crucial are roadside commercial activities, pavement condition
conclusions are drawn from the data analyses and field index, frequency of driveways carrying a high volume of
observations. The Pearson's correlation analysis carried out traffic, average traffic speed, and unauthorized on-street
for identifying the significant variables has reported that stoppages of intermittent public transits. Thus, taking ac-
eight several variables are contributing to the quality of tions for the minimization of street vending activities and
bicycling on urban road segments in mid-sized cities car- maintaining the smoothness of pavement surface, etc., are
rying heterogeneous traffic. These variables include effec- some other key factors for the improvement of bicycle ser-
tive width of the outermost lane (WOTLE), peak hour traffic vice quality.
volume per lane (PHVL), roadside commercial activities (CA), The study effort goes beyond the BLOS model development
pavement condition index (PCI), frequency of driveways by summarizing several crucial observations obtained from the
carrying a high volume of traffic (D), average traffic speed (S), model's field applications. As observed, around 95% of the
and unauthorized on-street stoppages of intermittent public studied segments are offering average (BLOS C) or inferior ser-
transits (IIPT). An in-depth review of the literature has vice qualities for the bicycle use. The performance of existing
revealed that, the regression analysis is being primarily used facilities may degrade over time period due to several reasons,
for the development of BLOS models. However, the present including a gradual increase in traffic volume, increase in
study has accomplished the BLOS model development commercial activities within urban areas and degradation in
through the application of a novel and advantageous pavement qualities, etc. Hence, the quality of existing facilities
modeling approach namely, functional networks (FNs). FN is should be timely augmented for the betterment of bicycle
one of the recently developed artificial intelligence tech- users. One major contribution of this research is the develop-
niques whose architecture is based on the structure of the ment of highly reliable and statistically calibrated BLOS model
physical world. Several added advantages have made the suitable for application in midsized cities under the influence of
technique as a powerful alternative to the artificial neural heterogeneous traffic conditions. The developed model and
networks (ANNs) approach. The BLOS model developed in other findings will help in long-term transportation planning,
this study has shown an excellent performance in the pre- assessment of on-street bicycling environment, development
sent context in terms of both goodness-of-fit and prediction of bicycle-friendly road infrastructures, identification of which
precision. The correlation coefficient (R) between model road attribute should be primarily prioritized to enhance the
predicted and overall perceived BLOS scores was observed to operational efficiency of the on-street bicycle facilities, identi-
be as high as 0.94. Several other statistical parameters, such fying most suitable bicycling route inside a city, and much
as NasheSutcliffe efficiency coefficient (E), error measuring more. As bicycle lanes are not available on study corridors, the
parameters, and overfitting ratio, etc., were also applied to developed model does not address the influence of this
model performance. As observed through this in-depth parameter, which is one limitation of this study. The model
investigation, the functional networks (FNs) based BLOS may not perform well in few specific locations where numerical
520 J. Traffic Transp. Eng. (Engl. Ed.) 2017; 4 (6): 503e521

measures of the road attributes are not within the attribute Florida Department of Transportation (FDOT), 2009. Quality/Level
ranges used in this study (Table 2). However, the model inputs of Service Handbook. Florida Department of Transportation,
in this study are collected from widely varying conditions (i.e., Tallahassee.
Gandomi, A.H., Yun, G.J., Alavi, A.H., 2013. An evolutionary
excellenteworst) and well represent the possible variabilities
approach for modeling of shear strength of RC deep beams.
and complexities persisting in roadway environments in mid- Materials and Structures 46 (12), 2109e2119.
sized cities. Thus, the model is well anticipated to perform Hallett, I., Luskin, D., Machemehl, R., 2006. Evaluation of On-
well in mid-cities of developing countries around the world. street Bicycle Facilities Added to Existing Roadways. FHWA/
The average traffic speed on urban roads of Indian mid-sized TXDOT-06/0-5157-1. Center for Transportation Research,
cities varies in a lower range of 23e46 km/h due to highly University of Texas, Austin.
heterogeneous traffic flow conditions. This probably limits Harkey, D.L., Reinfurt, D.W., Knuiman, M., et al., 1998.
Development of the bicycle compatibility index: a level of
the universal applicability of the model to some extent.
service concept. Transportation Research Record 1636, 13e20.
However, the study effort can easily be duplicated elsewhere, Hunter, W.W., Feaganes, J.R., Srinivasan, R., 2005. Wide curb lane
if required. conversions: the effect on bicycle and motor vehicle
interaction. Transportation Research Record 1939, 37e44.
Iodice, P., Senatore, A., 2012. Analysis of a Scooter Emission
references Behavior in Cold and Hot Conditions: Modelling and
Experimental Investigations. SAE Technical Paper 2012-01-
0881.
Iodice, P., Senatore, A., 2013. Road transport emission inventory
Abagnale, C., Cardone, M., Iodice, P., et al., 2014a. Analysis of a
in a regional area by using experimental two-wheelers
new measurement system of the chain strength for
emission factors. In: The World Congress on Engineering
electrically assisted bicycles. In: The American Society of
(WCE), London, 2013.
Mechanical Engineers (ASME) 12th Biennial Conference on
Iodice, P., Senatore, A., Meccariello, G., et al., 2010. Methodology
Engineering Systems Design and Analysis, Copenhagen, 2014.
for the analysis of a 4-stroke moped emission behaviour.
Abagnale, C., Cardone, M., Iodice, P., et al., 2014b. Theoretical and
SAE International Journal of Engines 2 (2), 617e626.
experimental evaluation of a chain strength measurement
Indian Road Congress (IRC), 1990. Guidelines for Capacity of
system for Pedelecs. Engineering Letters 22 (3), 102e108.
Urban Roads in Plain Areas. IRC, New Delhi.
Beura, S.K., Kumar, N.K., Bhuyan, P.K., 2016. Level of service for
Jensen, S.U., 2007. Pedestrian and bicycle level of service on
bicycle through movement at signalized intersections under
roadway segments. Transportation Research Record 2031,
heterogeneous traffic flow conditions. In: 12th Transportation
43e51.
Planning and Implementation Methodologies for Developing
Kang, K., Lee, K., 2012. Development of a bicycle level of service
Countries (TPMDC), Mumbai, 2016.
model from the user's perspective. KSCE Journal of Civil
Castillo, E., 1998. Functional networks. Neural Processing Letters 7
Engineering 16 (6), 1032e1039.
(3), 151e159.
Landis, B.W., 1994. Bicycle interaction hazard score: a theoretical
Castillo, E., Cobo, A., Gomez-Nesterkin, R., et al., 2000a. A general
model. Transportation Research Record 1438, 3e8.
framework for functional networks. Networks 35 (1), 70e82.
Landis, B.W., Vattikuti, V.R., Brannick, M.T., 1997. Real-time
Castillo, E., Cobo, A., Manuel, J., et al., 2000b. Functional networks:
human perceptions: toward a bicycle level of service.
a new network-based methodology. Computer-Aided Civil
Transportation Research Record 1578, 119e126.
and Infrastructure Engineering 15 (2), 90e106.
MathWorks, Inc., 2016. MATLAB User's Manual 9.0. MathWorks,
Castillo, E., Cobo, A., Gutierrez, J.M., et al., 1998. An Introduction
Inc., Natick.
to Functional Networks with Applications. Kluwer Academic
Nash, J.E., Sutcliffe, J.V., 1970. River flow forecasting through
Publishers, Boston.
conceptual models part I e a discussion of principles.
Castillo, E., Ruiz-Cobo, R., 1992. Functional Equations in Science
Journal of Hydrology 10 (3), 282e290.
and Engineering. Marcel Decker, New York.
Oke, O., Bhalla, K., Love, D.C., et al., 2015. Tracking global bicycle
Chellapilla, H., Beura, S.K., Bhuyan, P.K., 2016. Modeling bicycle
ownership patterns. Journal of Transport & Health 2 (4),
activity on multi-lane urban road segments in Indian
490e501.
context and prioritizing bicycle lane to enhance the
Sorton, A., Walsh, T., 1994. Bicycle stress level as a tool to
operational efficiency. In: 12th Transportation Planning and
evaluate urban and suburban bicycle compatibility.
Implementation Methodologies for Developing Countries
Transportation Research Record 1438, 17e24.
(TPMDC), Mumbai, 2016.
Transportation Research Board (TRB), 2000. Highway Capacity
Cochran, W.G., 1977. Sampling Techniques, third ed. John Wiley &
Manual 2000. Transportation Research Board, Washington DC.
Sons, New York.
TRB, 2010. Highway Capacity Manual 2010. Transportation
Davis, W.J., 1987. Bicycle Safety Evaluation (PhD thesis). Auburn
Research Board, Washington DC.
University, Chattanooga.
Turner, S., Shafer, S., Stewart, W., 1997. Bicycle Suitability Criteria
Davis, J., 1995. Bicycle test route evaluation for urban road conditions.
for State Roadways in Texas. TX-97/3988-S. Texas
In: Transportation Congress 1 and 2: Civil EngineersdKey to the
Transportation Institute, College Station.
World's Infrastructure, ASCE, San Diego, 1995.
Verma, M., Rahul, T.M., Reddy, P.V., et al., 2016. The factors
Epperson, B., 1994. Evaluating suitability of roadways for bicycle
influencing bicycling in the Bangalore city. Transportation
use: toward a cycling level-of-service standard. Transportation
Research Part A: Policy and Practice 89, 29e40.
Research Record 1438, 9e16.
J. Traffic Transp. Eng. (Engl. Ed.) 2017; 4 (6): 503e521 521

Mr. Sambit Kumar Beura is currently a PhD Dr. Prasanta Kumar Bhuyan is currently an
research scholar at Transportation Engi- assistant professor in the Department of Civil
neering Division, Department of Civil Engi- Engineering, NIT Rourkela, India. He has
neering, National Institute of Technology gained PhD from Indian Institute of Tech-
(NIT) Rourkela, India. He received his bach- nology (IIT) Bombay and postdoctoral
elor's degree in civil engineering from Orissa research experiences from Georgia Institute
Engineering College, Biju Patnaik University of Technology, USA. He has published over 40
of Technology (BPUT), Odisha, India in 2013. articles in international journals and confer-
His research activity is focused on capacity ence proceedings. He has guided 27 students
and level of service analysis of trans- for their master theses and guiding several
portation facilities, driver behavior, non- PhD scholars. His research activity is focused
motorized modes of transportation, urban transportation plan- on operational analysis of transportation facilities considering both
ning, soft computing and statistical modeling. motorized and non-motorized modes. Also he focuses on applica-
tion of GIS, GPS, statistical methods and artificial intelligence
techniques for transportation data analysis and modeling.

Das könnte Ihnen auch gefallen