Sie sind auf Seite 1von 8

2011 European Intelligence and Security Informatics Conference

Finding Criminal Attractors based on Offenders


Directionality of Crimes
Richard Frank, Martin A. Andresen
Connie Cheng, Patricia Brantingham
School of Criminology
School of Criminology
Simon Fraser University
Simon Fraser University
Burnaby, Canada
Burnaby, Canada
{rfrank, andresen}@sfu.ca
{cca25, pbrantin}@sfu.ca

Abstract According to Crime Pattern Theory, individuals all motivation, and others being crime attractors allowing for
have routine daily activities which require frequent travel many criminal opportunities that offenders are aware of and
between several nodes, with each being used for a different take advantage of [1]. Finally, other areas, termed criminal
purpose, such as home, work or shopping. As people move attractors, do not necessarily attract crime, but attract the
between these nodes, their familiarity with the spatial area criminals themselves, and they commit crimes along the way.
around the nodes, as well as between nodes, increases. Offenders
have the same spatial movement patterns and Awareness Spaces As a person wants to go from a node to another, they will
as regular people, hence according to theory an offender will not follow a random walk [2], but instead will attempt to travel
commit the crimes in their own Awareness Space. This idea is along the optimal route: the direct path, a straight line
used to predict the location of the nodes within the Awareness connecting the two nodes. However, due to boundaries, either
Space of offenders. The activities of 57,962 offenders who were natural, such as lakes and mountains, or man-made, such as
charged or charges were recommended against them were used highways and buildings, they cannot follow the direct path, and
to test this idea by mapping their offense locations with respect to instead must move around on circumscribed paths. In cities
their home locations to determine the directions they move. Once these circumscribed paths are usually along road-segments.
directionality to crime was established for each offender, a Since the movement is restricted by the topology of a persons
unique clustering technique, based on K-Means, was used to environment, the person will have a large set of fixed possible
calculate their Cardinal Directions through which the awareness paths to take from one node to the other, with the majority of
nodes for all offenders were calculated. It was found that, by
those paths being completely undesirable (leading away from
looking at the results of various clustering parameters, offenders
tend to move towards central shopping areas in a city, and
the destination with a significant detour, for example), leaving
commit crimes along the way. Almost all cluster centers were only a handful of reasonable paths. The person will travel along
within one kilometer of a shopping center. This technique of those select few routes that they consider optimal (or close to
finding Criminal Attractors allows for the reconstruction of the optimal). After the person has established a preference for
spatial profile of offenders, which allows for narrowing the certain routes over others, they will become familiar with that
possible suspects for new crimes. route, and it will be incorporated into their Awareness Space.
According to Crime Pattern Theory (CPT) [3], offenders act
Keywords-component; offender profile, directionality,
clustering, criminal attractor
on opportunities within the area they are familiar with and will
use those locations to commit crimes, rather than exploring
I. INTRODUCTION new areas with which they are not familiar with. This is
intuitively correct. If the offender leaves their Awareness Space
The movement of people is not random. Each person will in search of criminal opportunities, they will have to enter
routinely travel along paths between only a handful of unfamiliar territory and hunt for an opportunity; this exercise is
locations, such as their home, work and nearby shopping. With unnecessary if the offender is already aware of a similar
each and every trip, they will familiarize their knowledge opportunity within an area they are familiar with. The sub-
further about the path, and everything along that path. Because regions of the Awareness Space where the offender actually
people are familiar with the paths they travel frequently, they commits crimes make up the Activity Space. This phenomenon
are more comfortable in them, as opposed to moving into is shown in Figure 1.
unfamiliar areas. These familiar areas become the persons
Awareness Space. The difficulty with applying Crime Pattern Theory is that
the Awareness Space for an offender is very difficult to obtain
The Awareness Space is built from two components: nodes without direct access to the offender and their willingness to
and paths. The nodes, called Awareness Nodes, are formed share such information. Knowing the Awareness Space of an
from a persons frequent destinations, such as their work place, offender is important as it will highlight the possible Activity
residence or recreation, which provide the end-points of the Spaces of the offender, allowing law enforcement to harden
journey. From a criminological point of view, the Awareness those areas to prevent further offenses. Unfortunately without
Nodes are not created equal, with some being crime generators observing the offenders cognitive-map of an area, the
attracting a large number of people regardless of their criminal Awareness Space cannot be known, only the offenses

978-0-7695-4406-9/11 $26.00 2011 IEEE 86


DOI 10.1109/EISIC.2011.34
offenders were heading when they stopped to commit a crime.
Information about all Awareness Paths is then used to find the
areas (Nodes) within a city where offenders typically go as part
of their (normal non-criminogenic) lives. The eventual goal of
this model is to be able to profile offenders based on their
known spatial locations (such as their home) and their crime
locations. DCAL adds to the current literature through the
following contributions:
1) Using a popular clustering method in a novel way to
perform clustering of directions instead of spatial points.
2) Using clustering techniques to find where people are
going, based on data about where they offend.
3) Introducing a model capable of reconstructing the
Figure 1- Offenders Awareness Space and the Nodes within it
Awareness Space of each individual offender based on the
knowledge of many offenders known home and crime
committed by the offender (for which they have been caught) locations.
are known. Could the Awareness Space be determined by 4) Using real crime data to experimentally evaluate the
knowing a criminal subset of the Activity Space?
model and finding the Criminal Attractors of real offenders.
This question has significant meaning when profiling The paper first discusses the details of the model (Section
offenders. Profiling is a way of using the attributes of a crime, II), followed by an outline of the method through which it was
or crime location, to determine likely areas where the offender evaluated (Section III) to gather results (Section IV) on a real
could be. As mentioned above, crimes happen in areas known city in the Greater Vancouver Regional District in British
to offenders, which could be along the paths or close to the Columbia, Canada. The implications to criminology and
nodes in their Awareness Space. Knowing the location of the police of the results are discussed (Section V) and then the
crime, and the likely nodes in the neighbourhood, a possible paper is concluded (Section VI).
Activity Space could be determined. Projecting the Activity
Space into the Awareness Space could then highlight important II. THE MODEL
nodes in the Awareness Space of the offender, regions of space
Why do offenders offend at the locations they do?
where the offender could be living, for example. When a new
According to Crime Pattern Theory, the crimes an offender
crime occurs, it would be possible to narrow down the number
commits are going to be situated near a path in their Awareness
of suspects by finding offenders whose Awareness Space
Space, where the path leads from one node to another. So, how
intersects the crime location.
is the path structured? Where is its origin? Where does it
This not only application to offenders but also victims, as terminate?
they have the equivalent Awareness Spaces while they also
The above questions raise two underlying problems. First,
move about in their own environment. Further, studies have
for a highly prolific offender, there will be a large number of
also found that victims are victimized most often close to their
Journeys to Crime (JtC), navely one to each crime location. If
home, within their corresponding Awareness Spaces [4] [5].
multiple such JtC are headed in the same general direction,
One of the components of the Awareness Space, the called Cardinal Direction (CD), then these must be merged into
Awareness Path, is made up of two components: distance and a single path. The method proposed in this paper,
direction. The distance component of the journey to crime has Directionality-based Criminal Attractor Locator (DCAL), tests
been researched significantly over the years [6] [7] [8] [9] [10]. for, and detects, the need to merge multiple paths into a few
Research into Crime Pattern Theory (CPT) and the Journey to CDs (see Section II.A). The second problem is that the
Crime (JtC) has shown that offenders do not travel very far (known) crime location is not necessarily at the end of the
from their home location [11] [12] [13]. The distance to the awareness path, but could be anywhere along the path, thus the
crime location (from home) does tend to vary depending on location of the node at the end of the path remains unknown.
various factors, such as the crime type, age of offender(s), DCAL extends each individual path and uses clustering
layout of the city and the transportation routes available within techniques to find areas where a lot of paths lead to in order to
it [14] [15] [16] [17]. Although the distance component is well determine the locations where offenders are heading when they
researched, the directionality component of the paths has not commit crimes (See Section II.B). This allows DCAL to locate
achieved the same prestige and only few works exist that deal the nodes (that is, Criminal Attractors) in the Awareness Space
with it [18] [19]. for each individual offender, in effect reconstructing the
individual Awareness Spaces for all offenders.
In this paper, a Directionality-based Criminal Attractor
Locator (DCAL) model is proposed. DCAL exploits the A. Finding Cardinal Directions
directionality component of Awareness Paths in order to Multiple crimes could be the result of the offender
reconstruct the Awareness Space of offenders from their travelling along multiple paths in their Awareness Space.
Activity Space. It does so by attempting to predict the position However, prolific offenders could have multiple crimes along
and orientation of the Awareness Paths along which the the same general direction (Cardinal Direction), roughly

87
a) Individual Paths (one to each b) The corresponding Cardinal
crime location) Direction

Figure 2 Individual vs. Aggregate Paths Figure 3 Crimes spread out around home. No clustering spatially, but
clustered directionality-wise.

a) Mapping of crime locations to the unit circle b) Crime locations mapped to the unit circle c) Cluster centers and directions for the activity
paths identified.

Figure 4 Finding Activity Paths

heading towards the same node, (Figure 2a). It is not likely that Cardinal Directions requires direction-based clustering. DCAL
this offender has different Awareness Paths to each of their removes distance from the equation and focuses solely on the
criminal destinations, but rather that there is a single directionality component. This is done by taking each crime
Awareness path that goes near all of the offenses for that location and determining the exact distance between it and the
offender (Figure 2b). Thus, in general, instead of there being a central node, then (virtually) moving the crime location such
unique path for each offender, it is expected that some crimes that it maintains the exact direction with respect to home, but is
are the result of a single Cardinal Path. only a unit distance away (Figure 4a). The result is that all
crime locations are on the unit circle with the central node in
One possible solution to the problem is to simply look for the center (Figure 4b). Now, clusters based on directions are
clusters in space, where there is a spatial cluster, there is a apparent, and can be found (Figure 4c).
Cardinal Direction leading there. This approach however is not
correct. It is possible that the crime locations are in a similar DCAL finds these clusters, and their cluster-center, using
direction, but spread out in space, and hence do not form a the popular K-Means clustering algorithm [20]. Since K-Means
spatial cluster. In Figure 3, for example, a spatial clustering requires the number n of clusters to be specified by the user as
algorithm would most likely detect approximately five spatial an input parameter, and the number of clusters is not known a
clusters (shown in circles) although it is clear that those five priori, thus a large number of clusterings are performed with an
spatial clusters are the result of only two Cardinal Directions iteratively larger number for n. The clustering is stopped when
(shown as arrows). Additionally, the four crimes towards the the error rate of the clustering drops below 0.1, at which point
lower right do not make up a spatial cluster, but they are the the clustering is assumed to be good with n clusters. Clusters of
result of a single Cardinal Direction. The reason why spatial one are assumed to be outliers and are dropped1. The remaining
clustering is not a valid approach in detecting Cardinal
Directions is because the different clusters (referring to Figure 1
Clusters of one would indicate a path that is not frequently travelled (at least
3) are at various distances away from the central node. Spatial from a criminological point of view), and hence do not form a Cardinal
clustering is implicitly distance-based, whereas finding Direction. The model works identically if these data-points are kept.

88
a) Establish direction of paths b) Extend Cardinal Direction to find intersections

c) Finding Awareness Nodes by looking for clusters of intersection points d) Identify offenders Awareness Space and Paths

Figure 5 Finding Awareness Nodes and Paths to find the Awareness Space

cluster centers are then used as the directions for the CDs. In goal is to determine where the offenders are moving to, their
the example shown in Figure 4c, there would be three CDs: one CDs are extended an arbitrary distance beyond their crime
towards the upper left, upper right and lower right. A singleton location, and are intersected with the CDs of other offenders
cluster towards the lower left is removed as an outlier. (Figure 5b). The idea is that the spatial region with a lot of
intersections will have a lot of offenders moving towards it,
Due to tractability purposes, a single restriction must be implying that that region is of interest. Once all intersections
placed onto the initial model presented in this paper: each path are calculated, K-Means [20] is applied to the set of
is assumed to start at the home of the offender (and not one of intersection points (Figure 5c). Once again, the appropriate
the nodes in their Awareness Space). number of clusters n is unknown, hence multiple clusterings are
B. Finding Nodes in the Awareness Space performed while varying n, with the idea that regions of space
According to theory, each path in the Awareness Space is which consistently contain cluster centers under different
created by frequently travelling between two nodes, and clusterings are stable, and are the true areas of interest. With
committing crimes somewhere in between. In other words, the stable cluster centers detected, offenders who have a Cardinal
JtC is not necessarily as long as the journey to the Criminal Direction pointing towards one of those cluster centers have
Attractor, but that it is in the same direction with respect to the that node as part of their Awareness Space (Figure 5d).
starting point. This implies that the node in the Awareness III. EXPERIMENTAL SETUP
Space is going to be found by continuing along the JtC an
arbitrary distance beyond the crime location. The challenge is To evaluate DCAL, the model was implemented in Visual
to determine how far beyond it, as the theory does not state Studio 2008, with the data retrieved from DB2, and the
where along the path the crime is, only that it is along the path. clustering process was passed over to Matlab 2009as K-
Means implementation. The model was then applied to repeat
It is assumed that certain areas will act as the node to a lot offenders2 residing in the city of Surrey, a part of the Greater
of individuals, thus, to determine where the node is beyond the
Vancouver Regional District in the South-West corner of
crime location, the model uses the Cardinal Directions of all
British Columbia, Canada. This city was picked specifically
offenders in the region to see which spatial area a lot of
offenders are moving towards. This is done as follows. for this model, since visual analysis of the citys offender
residence and crime locations did not yield any patterns, but
As in the previous step, the Cardinal Directions are found only chaotic movement. At the time, it was thought that the
for each offender, after which the focus becomes using only the
CDs for all offenders simultaneously (Figure 5a). Since the 2
A repeat offender was considered to have at least 5 offenses.

89
single-detached, according to the results of the 2006 Canadian
Census Survey [25]. Average household income within the city
was approximately $73,000, but varied within the city from
$45,000 (in Whalley) to over $96,000 (in South Surrey) [25]. It
is connected to near-by cities to the North and West, and
through them to Vancouver, by several bridges. The
community of South Surrey is adjacent to the border separating
Canada from the United States.
B. Data
For the analysis presented below, a collection of databases,
called the Crime Data-Warehouse (CDW) was used. The CDW
a) Data outside the edges b) No data outside the edges is a research database housed at the Institute of Canadian Urban
Research Studies (ICURS) at Simon Fraser University. It
Figure 6 Cluster centers move inwards if data from edges are eliminated. contains five years of real-world crime data retrieved for the
Province of British Columbia (BC), Canada, from Canadas
citys grid patterned road-network was distorting any apparent national police, the Royal Canadian Mounted Police (RCMP).
patterns beyond visual recognition and this made a good test The RCMP in BC uses a system called the Police Information
for an automatic process such as DCAL. Retrieval System (PIRS), a large database keeping information
Below is a discussion of Surrey, the city used for the study for the regions of the province of BC which is policed by the
(Section III.A) and the data used for the evaluation (Section RCMP. PIRS contains information about crime-events
III.B). Note that although the experimental evaluation focused (~4.4million) and people (~9million), in addition to vehicles
on the city of Surrey, crime and offender data about Surrey (~1.4million) and businesses (~1.1million). Crime-events will
and all neighbouring jurisdictions were entered into the model. hereafter be called simply crimes.
This was due to the edge-effect [21] [22] where patterns near For this study, only the crimes and the people associated to
the edges are distorted due to a lack of data on the other side those specific crimes were studied. Information about vehicles
of the edge. This causes results near the edges to be and businesses were not used. The data was restricted to all
underrepresented. Certain types of clustering algorithms, K- crimes occurring between August 1, 2001 and July 31, 2006,
Means being one of them, are particularly vulnerable to this committed by offenders within the study area. Linkages exist in
phenomenon. More specifically, the result is a tendency for the database between the crimes and people table, thus all the
the cluster centers to move inwards towards the center of the people involved in any given crime were identified. This paper
dataset if data are removed from the edges, as shown in Figure focuses only on property crimes, and for only those offenders
6. Thus, in order to displace the edge effect to the outside of who had at least 5 offenses for which they were charged, in the
the study area, data for neighbouring cities was included. process of being charged, had charges recommended against
Thus, although data outside of Surrey is part of the model, the them, or were suspects in the offense.
results are only considered valid within Surrey, and not in any For the people involved in each crime, the relevant
other city, where its considered unreliable because of attributes for this model include the full name (first, last and
distortions in the clustering results due to a lack of data middle), home address and the type of their involvement in the
immediately outside their respective boundaries. crime. People having the same name were assumed to be the
same person. Linked to each person was the set of crimes,
A. Study Area
along with the type and location of the crime, that they were
As mentioned above, the city of Surrey, a part of the involved in. If the location specified in the database was invalid
Greater Vancouver Regional District in the South-West corner and could not be assigned an XY coordinate, the location was
of British Columbia, Canada, was chosen as the locale for ignored for the analysis. Although for this specific study no
evaluating DCAL due to the chaotic movement patterns
other information about offenders was available, such as the
observed visually within the city. Although the citys road-
network is a grid-network, the crime locations did not at all location of their work or other activities, this type of data could
match expectations during an initial visual analysis. have been used to enhance the models accuracy.
For the experiments conducted for this paper, it was
The city, established in 1879, is the second largest city assumed that each JtC started at the home location, that is,
(after Vancouver) in the Province of British Columbia with a each path is assumed to start at the home address of the
population of approximately 400,000 [23]. In 2005, Surrey was offender (and not one of the nodes in their Awareness Space).
named Canadas worst city for car-theft, with a reported 7,654
car-theft related calls-for-service to the police3, a decrease of IV. RESULTS
8.2% from the previous year [24]. The city is a suburb of
The home and crime locations of all offenders living in
Vancouver, and thus almost half of all housing in Surrey is
Surrey and its surrounding cities were input into DCAL. This
resulted in 57,962 repeat offenders (spread across all cities).
3
This number included cars recovered in the city of Surrey, but stolen The CDs for each individual offender were found, with most
elsewhere, along with those cars that were stolen from within the borders of of the offenders having 3 CDs. The CDs were extended and
Surrey.

90
Figure 7 Cluster centers in Surrey

a) Central City Shopping Center b) Guildford Town Center c) Center of Newton

Figure 8 Neighbourhood of the cluster groups. M denotes a Shopping Mall, of which there are only three in Surrey.

the intersections calculated to yield approximately 10 million unreliable. This includes the small cluster-group towards the
intersections, which were then clustered. Since it was not West of Surrey, and the few isolates in the North and East.
known a priori how many clusters were to be found in Surrey, Removing those, there are three clear cohesive cluster-groups
clustering was done many times, with K-Means being told to within Surrey (shown in Figure 8).
find one to ten cluster centers in each iteration. Areas with a The area around the cluster-group towards the North-West
lot of cluster centers were assumed to show areas of high of Surrey, shown in Figure 8a, is a major hub within the city,
interest. The results are shown in Figure 7. and contains one of Surreys largest shopping malls (Central
As stated above, cities outside of Surrey were included in City - identified with an M in the figure), a campus of Simon
the dataset to eliminate the edge-effect from within Surrey, Fraser University, two Skytrain stations, some government
thus all results (cluster-centers) outside of Surrey are deemed offices and tower complexes, along with major commercial

91
blocks nearby. Within the group, all but two of the cluster categorized as both crime generators and crime attractor where
centers are immediately adjacent to the location of the they are not only bringing in a large amount of people in
shopping center, while the other two cluster-centers are within particular places and times for crime to occur, but are also
one kilometer along the road-network. well-known places for criminal opportunities. Thus, other than
Another cluster-group, towards the North-East of Surrey, the explanation that offenders commit their crime on their way
shown in Figure 8b, is a bit more spread out. One cluster- to their daily activity, it is also possible that some offenders
center falls right on top of Guildford Town Center, Surreys intended to commit offenses in these malls but see the
largest shopping mall. While the other cluster-centers are opportunity during their journey to the targeted location.
somewhat removed from the shopping district, they fall onto Another interesting result is that by looking at the cluster-
several Secondary Schools in the area. The furthest cluster- center, we may be able to determine some of the
center is approximately two kilometers away (along the road- characteristics of the offenders who commit crime in that
network) from the mall and is located at a Secondary School. particular area, narrowing down possible suspects. For
It is possible that offenders whose direction is pointing toward instance, the cluster center on the right in Figure 8b is in
these locations may frequently commute to either or both of between a school and a shopping mall. It is possible that
these nodes, perhaps even as students of the Secondary offenders whose direction points to this cluster center have a
School. Otherwise the entire region is mainly low-density relationship with the school. They could be students, workers,
residential housing. or parents from the school. By understanding the relationship
The final cluster-group to the South, Figure 8c, is again between offenders choice of direction and the nature of the
near a shopping center (Center of Newton) with two blocks of places, it may be able to help the police on their investigation
commercial areas and a hospital surrounding it. Four of the by providing a profile of the offenders.
cluster-centers are within a few hundred meters of the The most benefit of this type of spatial profiling could
shopping center, and the furthest cluster-center from Center of come when a new crime event occurs. Given that the police
Newton only a block away, approximately 500 meters. have data similar to what was used for this paper, they could
There was a group of four cluster-centers at the far right of use the process proposed in this paper to reconstruct the
Figure 7, seemingly in the water, near an industrial area Awareness Spaces of all offenders in their database. When a
without anything significant near it. Although this cluster- new crime event happens, according to theory, it would occur
group is cohesive, like the other three, it is likely that it is in the Awareness Space of an offender, thus, the police could
distorted due to the edge-effect and thus not in the proper see which offender has an Awareness Space intersecting the
location. If more data had been used in the cities outside of location of the new crime event, and investigate those as the
Surrey, those cluster-centers would most likely have moved initial suspects. If none of the offenders are responsible, the
slightly further out from Surrey, right into downtown New method from this paper could still yield valuable information
Westminster, which is located immediately one kilometer by letting the police see the crime patterns, and home
north of the cluster-group. locations, of other offenders who have also committed a crime
in that region. Additionally, it could reveal possible other
V. DISCUSSION nodes this new offender could frequent. For example, given
Interestingly, all three cluster-groups were in close that the police know of three attractors in Surrey (A, B and C
proximity to one of the shopping facilities in Surrey. Although
there are two other shopping centers in Surrey, those did not
seem to create any cluster-centers at all, regardless of the
number of clusters sought by K-Means. The reason for this is
unknown, but could be that they do not have certain features
which attract offenders, that the other three shopping centers
have. Further investigation into this phenomenon is required.
The cluster-centers were built from the CDs of offenders,
implying that most offenders moved toward a shopping center
when they commit crimes. According to Crime Pattern Theory
[3], Awareness Space is built around peoples activity nodes.
By looking at the cluster centers, it is clear that the major
shopping centers in Surrey are an integral part of offenders
activity nodes. Because the cluster-group on the North-West
side of Surrey (near Central City) is surrounded by two
Skytrain Stations, the stations themselves may be one of the
activity nodes offenders frequent, perhaps in their commute to
downtown Vancouver, to which there is a direct Skytrain line.
It is interesting that most offenders commit their crimes
along their journey to a shopping center. According to the idea Figure 9 Using DCAL for profiling (star indicates new crime location)
of crime generators and attractors [1], shopping malls can be

92
on Figure 9), for any new crime the police could determine the volume 5, edited by Ronald Clarke and Marcus Felson. New
Brunswick, NJ: Transaction Publishers.
Awareness Nodes of offenders who have also committed
4) Caywood, T. (1998). Routine activities and urban homicide: A tale of
crimes near the new crime location (node C, for example) and two cities. Homicide Studies, 2(1), 6482.
then trace the path back from the node, through the crime, to 5) Groff, E., & McEwen, T. (2006). Exploring the spatial configuration of
the possible area where the offender could be living (shaded places related to homicide events: Final report. Washington, DC:
region on Figure 9). National Institute of Justice.
6) Brantingham, P.J., and Brantingham, P.L. (1984), Patterns in Crime.
VI. CONCLUSIONS New York, NY: Macmillan.
This paper presented an approach to find the Awareness 7) Ratcliffe, J.H. (2006), A temporal constraint theory to explain
opportunity-based spatial offending patterns, Journal of Research in
Space of offenders by using Crime Pattern Theory to predict Crime and Delinquency, 43: 261 291.
each individual offenders Awareness Space, based on 8) Scellato, S., Cardillo, A., Latora, V., and Porta, S. (2006), The
knowledge gained from the patterns of all offenders. This was backbone of a city, European Physical Journal B - Condensed Matter
done by applying clustering techniques in a new way to find and Complex Systems, 50: 221 225.
the Cardinal Directions of each individual offender in order to 9) Groff, E.R. (2007), Situating simulation to model human spatio-
determine the directions that they tend to go to commit crimes. temporal interactions: an example using crime events, Transactions in
Given that a crime could occur anywhere on a path, not GIS, 11: 507 530.
necessarily at either end of it, the Cardinal Directions were 10) Groff, E.R., and McEwen, T. (2007), Integrating distance into mobility
extended and intersected with each other with the idea that triangle typologies, Social Science Computer Review, 25: 210 238.
intersections represent points where two people are moving 11) Wiles, P., and Costello, A. (2000), The road to nowhere: the evidence
for traveling criminals. Home Office Research Study 207. London:
towards, thus, areas which contain a lot of intersections Home Office, Research, Development and Statistics Directorate.
represent areas where a lot of people want to move towards. K- 12) Rossmo, Kim (2000). Geographic Profiling. Boca Raton, FL: CRC
Means clustering was once again applied to determine areas of Press.
high concentrations of intersections, that is, the nodes in the 13) Felson, Marcus (2002). Crime and Everyday Life. Thousand Oaks, CA:
Awareness Space. Given these nodes, the profiles of the Sage.
offenders can be reconstructed, if an offender has a crime in the 14) Phillips, P. D. (1980), Characteristics and typology of the journey to
direction of a node, they were most likely travelling towards crime, in K.D. Harries, ed., Crime: A Spatial Perspective, 167 180.
the node when they committed their crime. New York, NY: Columbia University Press.
15) Castanzo, C.M., Halperin, W.C., and Gale, N. (1986), Criminal
The activities of 57,962 offenders who were charged or mobility and the directional component in journeys to crime, in R.M.
charges were recommended against them were used to test this Figlio, S. Hakim, and G.F. Rengert, eds., Metropolitan Crime Patterns,
idea by mapping their offense locations with respect to their 73 95. Monsey, NY: Criminal Justice Press.
home locations to determine the directions they move in. It was 16) LeBeau, J. L. (1987), The journey to rape: geographical distance and
found that, by looking at the results of various clustering the rapists method of approaching the victim, Journal of Political
Science and Administration, 15: 129 136.
parameters, offenders tended to move towards central shopping
17) Van Koppen, P.J., and De Keijser, J.W. (1997), Desisting distance
areas in a city, and commit crimes along the way. Almost all decay: on the aggregation of individual crime trips, Criminology, 35:
cluster centers were within one kilometer of a shopping center. 505 515.
This finding allows for the reconstruction of the spatial profile 18) Brantingham, P.L., and Brantingham. P.J. (1981), Notes on the
of offenders, which will allow for narrowing the possible geometry of crime, in P.J. Brantingham and P.L. Brantingham, eds.,
suspects to new crimes. Environmental Criminology, 27 53. Beverly Hills, CA: Sage
Publications.
The model does have some shortcomings, which will be the 19) Rengert, G.F., and Wasilchick, J. (1985), Suburban Burglary: A Time
investigated in the continuation of this project. First, the edge- and a Place for Everything. Springfield, IL: Charles C. Thomas.
effect is biasing results near the edge of the city. Even though 20) Kanungo, T.; Mount, D. M.; Netanyahu, N. S.; Piatko, C. D.;
data from the surrounding cities were included, it did not seem Silverman, R.; Wu, A. Y. (2002). "An efficient k-means clustering
enough to completely eliminate its effect from within Surrey. algorithm: Analysis and implementation". IEEE Trans. Pattern
Analysis and Machine Intelligence 24: 881892.
Second, the results presented in this paper are for a single city
21) Brantingham, P.L. and Brantingham, P.J. Nodes, paths and edges:
within the GVRD. The approach and experiments should be Considerations on the complexity of crime and the physical
repeated on several other cities to understand the patterns and environment, Journal of Environmental Psychology, Volume 13, Issue
attractors each city has. Finally, an entire profiler system 1, March 1993, Pages 3-28
should be built which incorporates the techniques from this 22) Frank, R., Park, A.J., Brantingham, P., Clare, J., Wuschke, K. and
paper, to actively profile offenders in order to help police. Vajihollahi, M Identifying High Risk Crime Areas using Topology
IEEE Intelligence and Security Informatics
VII. REFERENCES 23) Statscan.ca "2006 Community Profiles - Census Subdivision".
1) Brantingham, P.J. and Brantingham, P.L. (1995). "Criminality of Place: 2.statcan.ca. 2010-12-07. Retrieved 2011-05-01.
Crime Generators and Crime Attractors". European Journal on Criminal 24) The Vancouver Sun We're No. 1, but it's a dubious honour,
Policy and Research 3(3):1-26. November 15, 2005. Retrieved 2011-05-01.
2) Barber, M. N. and Ninham, B. W. (1970). Random and Restricted http://www.canada.com/vancouver/vancouversun/news/editorial/story.h
Walks: Theory and Applications. New York: Gordon and Breach. tml?id=890ff728-599b-4238-81d4-daa7df8f3204
3) Brantingham, P.J. and Brantingham, P.L. (1993). Environment, 25) City of Surrey, Community Demographic Profiles, 2006. Retrieved
on 2011-05-02 from http://www.surrey.ca/for-business/1417.aspx.
Routine, and Situation: Toward a Pattern Theory of Crime. Routine
Activity and Rational Choice, Advances in Criminological Theory,

93

Das könnte Ihnen auch gefallen