Beruflich Dokumente
Kultur Dokumente
Received: 18 September 2013 / Accepted: 31 October 2013 / Published online: 19 November 2013
# Springer Science+Business Media New York 2013
Abstract In this study, different anthocyanin sources including grape skin, black carrot, and red cabbage were used to
determine the effect of thermal treatment, different acidity
levels, and time on the anthocyanin content and degradation.
The total anthocyanin contents were modeled by neuro fuzzy
inference system (ANFIS) and artificial neural network
(ANN) models. The red cabbage anthocyanin stabilities were
higher than others. The anthocyanins degraded more rapidly
at higher temperatures. The anthocyanin contents of samples
decreased with the increase of pH from 3 to 7. Comparison of
the models showed that the ANFIS model performed better
than the ANN model for the estimation of total anthocyanin
content in all samples. The lowest root mean square error
(0.0457) and highest R 2 (0.9942) values were obtained for
red cabbage and grape skin in the validation period with the
ANFIS model, respectively. This study showed that both
models can be utilized efficiently for the prediction of total
anthocyanin content affected by temperature, time, and pH.
Introduction
Interest in anthocyanins as natural colorants and antioxidant
ingredients has recently increased due to their color characteristics and potential health benefits (Walkowiak-Tomczak
and Czapski 2007). Furthermore, epidemiological studies
have suggested that the anthocyanins have a protective effect
against some cancers (Castaneda-Ovando et al. 2009).
Anthocyanins are glycosides of anthocyanidins (aglycone),
which differ from one another in the position of substitution of
hydroxyl and methoxy groups in the -ring. Anthocyanin
residues are frequently acylated with phenolic acids
(Walkowiak-Tomczak and Czapski 2007). Some factors like
pH, temperature, copigments, light, metallic ions, enzymes,
oxygen, ascorbic acid, sugar, and their degradation products
can cause anthocyanin degradation and deterioration of food
color (Mazza and Miniati 1993).
Black carrot, red cabbage, red radish, red potato, and grape
pomace are popular anthocyanin sources (Newsome 1986).
For example, the black carrot (Daucus carota L. ssp. sativus
var. atrorubens Alef.) originating from Turkey, Afghanistan,
Egypt, Pakistan, and India has been known for 3,000 years
(Kammerer et al. 2004). Red cabbage (Brassica oleracea L.
var. capitata f. rubra) which belongs to the family of Brassicaceae grows all over the world (Arapitsas et al. 2008). The
grape (Vitis vinifera L.) has an annual production rate of
around 69 million tons and is cultivated mainly for wine
production (Maier et al. 2009). Wine industry wastes may
account almost 30 % (w /w ) of the grapes used in wine
production. Varietal differences and wine-making processes
are the major factors which can change the phenolic content
and composition of grape pomace (Ruberto et al. 2007). In this
research, the skins of one of the most important red wine grape
varieties in Turkey Okuzgozu, black carrot, and red cabbage
were used as anthocyanin sources. The stability of the anthocyanins is closely related to a set of common physicochemical
1329
term Gaussian membership functions exhibited the best results for antioxidant activity and anthocyanin content, respectively (Taghadomi-Saberi et al. 2013b).
The aim of the present study was firstly to determine the
stability of the anthocyanins obtained from black carrot, red
cabbage, and grape skin in buffer systems, at different temperatures and acidity levels, and secondly to predict the total
anthocyanin content of extracts depending on temperature,
time, and pH using ANFIS and ANN modeling techniques.
1330
different pH levels (3.0, 4.0, 5.0, 6.0, and 7.0) at 70, 80, and
90 C for 120 min. The citratephosphate buffer solutions at
these pH levels were colored with anthocyanin-based extracts
(100-ml buffer solutions were colored with 4 mg anthocyanin;
black carrot extract, 1.20 g; red cabbage extract, 0.50 g; grape
skin extract, 0.37 g), and they were divided into 15-ml portions and poured into Pyrex tubes and incubated in a thermostatic water bath (Memmert WB-22, Germany) which has
been preheated to the given temperature. Samples were removed from the water bath at 30-min intervals and cooled by
running tap water. The anthocyanin contents were measured
by the pH differential method described by Giusti and
Wrolstad (2001).
Total Anthocyanin Content
Determination of the total anthocyanin contents of the samples
was performed according to the pH differential method (Giusti
and Wrolstad 2001), and values were expressed as cyanidin-3glucoside (cyd-3-glc) equivalents, according to the following
equation: c (mg/kg)=(A MWDF1,000d)/ M, with A =
absorption value, MW = molecular weight (449.2 g/mol for
cyd-3-glc), M = molar extinction coefficient (26,900 L/mol/
cm for cyd-3-glc), DF = dilution factor, and d = path length of
the cuvette (1 cm). Absorbance of the sample was determined
using a spectrophotometer (Varian Cary 100 Conc UVVisible, USA).
Adaptive ANFIS Model
The fuzzy logic toolbox of MATLAB 7.0.1 was used for
ANFIS modeling (MATLAB 2009). Firstly, the experimental
parameters of the samples were classified. In the construction
of a set of fuzzy if-then rules with appropriate membership
functions (MFs) from the specified inputoutput pairs, ANFIS
employs an ANN learning algorithm. The two methods for
updating MFs used by ANFIS in the training phase are (a)
back propagation for all parameters (a steepest descent method) and (b) a hybrid method consisting of back propagation
for the parameters associated with the input membership and
least squares estimation for the parameters associated with the
output MF (Jang 1993). Figure 1a illustrates the general
architecture of an ANFIS model. A fuzzy inference system
(FIS) is composed of two inputs x and y and one output z. For
the first-order Sugeno fuzzy model, a typical rule set with two
fuzzy if-then rules can be expressed as:
Rule 1 If x is A1 and y is B1, then z i =p 1x +q 1y +r 1
Rule 2 If x is A2 and y is B2, then z 2 =p 2x +q 2y +r 2
where the A1 and A2 and B1 and B2 are the MFs for inputs x
and y, respectively, and p i, q i, and r i are the output functions
parameters. The ANFIS system is functionally equivalent to
MAE
N
1 X
Y iobserved Y ipredicted
N i1
1331
Input
Inputmf
Rule
Outputmf
Output
Temperature
pH
Anthocyanin
Time
b
Statistical Analyses
Each experiment was repeated at least two times in triplicate.
SAS statistical software (SAS 2000) was used for the data
analysis, and all the tests were considered to be significant if P
<0.05.
1332
Table 1 Thermal degradation of black carrot, red cabbage, and grape skin anthocyanins (milligram cyd-3-glc/L) at different temperatures, times, and pH
levels
pH Time (min)
Black carrota
Red cabbage
Grape skin
0
30
60
70 C
40.050.59
38.750.87
38.440.28
80 C
40.040.45
38.030.59
37.150.96
90 C
40.030.53
32.251.52
28.011.30
70 C
40.030.59
39.700.30
39.500.47
80 C
40.060.37
39.490.26
38.740.65
90 C
40.050.41
35.510.92
32.910.78
70 C
40.020.40
38.250.22
37.460.80
80 C
40.060.58
38.010.49
37.260.60
90 C
40.050.55
31.120.85
26.731.34
90
120
Degradation %
0
30
60
90
120
Degradation %
0
30
60
90
120
Degradation %
0
30
60
37.930.89
36.300.67
9.36
40.030.52
38.371.48
36.800.50
36.321.39
35.241.09
11.97
40.030.32
38.131.10
36.700.89
35.821.62
34.510.81
13.79
40.021.17
37.251.65
34.951.47
36.180.85
35.361.05
11.69
40.040.50
36.821.20
34.750.81
33.120.78
31.721.35
20.78
40.050.75
34.200.47
32.361.13
29.390.87
27.591.16
31.11
40.000.54
32.870.52
29.040.85
24.992.16
22.162.34
44.64
40.020.45
31.871.37
28.110.97
25.220.95
22.660.74
43.38
40.030.57
30.821.68
27.480.94
24.490.79
21.190.58
47.06
40.020.64
28.421.48
22.681.38
39.300.37
39.000.42
2.57
40.030.47
38.710.73
38.470.31
38.170.56
37.440.72
6.47
40.010.46
38.580.34
38.150.41
37.870.60
37.410.61
6.50
40.030.79
38.450.80
38.150.23
38.250.61
38.040.63
5.04
40.050.70
37.150.80
36.410.91
35.750.93
35.410.72
11.59
40.020.26
35.820.71
33.210.32
31.080.38
30.160.88
24.64
40.030.37
34.201.28
32.940.62
31.450.98
29.600.60
26.09
40.030.46
34.790.78
31.910.81
30.610.62
28.700.96
28.30
40.050.30
32.720.69
30.440.96
27.370.47
26.701.34
33.33
40.060.35
31.140.38
27.860.58
36.620.71
35.290.80
11.82
40.050.38
37.800.48
36.820.18
35.841.01
35.020.68
12.56
40.010.48
36.470.44
35.050.58
32.841.02
31.680.31
20.82
40.060.55
36.241.07
32.400.96
36.320.49
35.110.49
12.36
40.040.54
37.210.75
35.380.28
33.120.52
31.980.36
20.13
40.020.34
31.400.34
28.750.38
26.220.71
24.970.18
37.61
40.070.41
28.250.30
24.300.41
23.280.89
20.110.79
49.79
40.090.63
30.630.90
27.000.85
24.170.97
21.900.22
45.37
40.100.65
29.740.63
24.940.59
22.200.61
20.470.92
48.95
40.070.71
19.481.26
14.050.30
90
120
Degradation %
0
30
60
90
120
Degradation %
33.861.90
30.570.89
23.61
40.020.60
32.701.12
29.151.31
27.110.92
25.771.55
35.61
25.920.51
22.770.73
43.08
40.041.08
31.131.79
27.731.56
24.400.95
22.090.86
44.83
17.531.13
14.490.46
63.79
40.000.65
26.201.53
20.951.18
16.320.46
14.301.83
64.25
37.770.62
37.250.67
6.94
40.060.37
37.301.04
35.790.31
34.890.32
34.050.45
15.00
32.100.55
31.170.35
22.13
40.010.53
31.590.58
27.910.63
25.680.31
23.790.61
40.54
25.090.85
23.261.17
41.94
40.050.43
28.551.19
26.070.62
21.210.69
19.250.93
51.94
30.921.00
28.810.25
28.08
40.040.49
28.870.71
24.690.63
22.250.50
20.100.52
49.80
21.940.72
19.290.82
51.86
39.960.82
23.711.15
16.810.60
14.460.39
12.050.64
69.84
11.970.42
9.971.40
75.12
40.090.58
19.121.09
13.490.65
9.950.74
7.280.71
81.84
study, Cevallos-Casals and Cisneros-Zevallos (2004) investigated the heat and pH stability of red sweet potato, purple
corn, and commercial purple carrot colorants and reported that
acylated anthocyanin sources were more stable than nonacylated ones. Red cabbage anthocyanins had the highest heat
stability because of their unique anthocyanin composition. In
particular, the highest heat stability of red cabbage anthocyanins is based on their acylated anthocyanin compositions
(Cevallos-Casals and Cisneros-Zevallos 2004; Dyrby et al.
2001; Pliszka et al. 2009). Cyanidin, which occurs as
cyanidin-3-sophoroside-5-glucoside and cyanidin 3,5diglucoside, acylated with sinapic, malonic, ferulic, and pcoumaric acids is the major aglycone of red cabbage
(Walkowiak-Tomczak and Czapski 2007). On the contrary,
the majority of anthocyanin sources such as grape skin anthocyanins which are all monosides and less acylated show light
resistance to heat, light, and pH changes (Hong and Wrolstad
1990). Black carrot anthocyanin pigment which is acylated
with p -coumaric, ferulic, p -hydroxybenzoic, and sinapic
acids exhibits higher hydration, light, and pH stability
(Kammerer et al. 2004).
The color stability of anthocyanins is attributed to pH and
their anthocyanin structures (Dai et al. 2009). Dyrby et al.
(2001) reported the heat stability of red cabbage, blackcurrant,
elderberry, and black grape colorants at pH 3 and at 80 C to be
red cabbage > blackcurrant > grape skin > elderberry. CevallosCasals and Cisneros-Zevallos (2004) found the order of stability of some anthocyanin sources to be red sweet potato > purple
carrot > purple corn > red grape at pH range of 0.9 and 4. These
results indicate that anthocyanin compositions contribute to
stability and that acylated anthocyanins are more resistant than
the non-acylated anthocyanins.
xi xmin
b
xmax xmin
ANFIS
Predicted Anthocyanin Values
where x min and x max are the minimum and maximum of the
training data set. The a and b were taken as 0.6 and 0.2 in this
study, respectively. The data were scaled between 0.2 and 0.8
and it was observed that this follows the suggestion of
Cigizoglu (2003) who demonstrated that scaling input data
between 0.2 and 0.8 provide flexibility for the models to
predict beyond the training range. Then, different ANFIS
and ANN architectures were tried to find the appropriate
model structures. Gauss MFs were used for the ANFIS in
the current study. In each application, different numbers of
MFs were tested and the best one which results the minimum
RMSE was selected. The best results for the ANFIS models
were recorded with the model which has seven membership
functions. The ANN with one hidden layer was used and the
trial and error method was applied to determine the hidden
nodes number. Sigmoid and linear activation functions were
used for the hidden and output nodes of the ANN, respectively. The training of ANN networks was stopped after 100
epochs because the variation of error was too small after this
epoch.
Table 1 shows the comparison of the training, testing, and
validation results of the ANFIS and ANN models. The optimal number of MFs of each ANFIS model and nodes in the
input, hidden, and output layers of each ANN model is also
tabulated in this table. As can be clearly seen from the table,
the ANFIS model performance was determined to be much
better compared to the ANN models for the training, testing,
and validation periods. The ANFIS model estimated the anthocyanin content of the samples with a quite high degree of
accuracy (RMSE<0.0457, MAE<0.5268, and R 2 >0.9942).
50
40
30
20
10
0
0
10
20
30
40
50
40
50
50
ANN
Predicted Anthocyanin Values
1333
40
30
20
10
0
0
10
20
30
1334
ANN model (Table 2). The ANFIS model showed a rather high
determination of the coefficient (R 2) for validation period
compared to the ANN for all the samples. While the R 2 values
for black carrot, grape skin, and red cabbage were determined
to be 0.9769, 0.9942, and 0.9836, respectively, for the ANFIS
model, they were calculated to be 0.9674, 0.9809, and 0.9683,
respectively, using the ANN model. A similar trend was observed for the MAE values. The MAE values for the black
carrot, grape skin, and red cabbage were calculated to be
0.8674, 0.5491, and 0.5268 in the ANFIS model, while they
were 1.0359, 0.9539, and 0.7257, respectively, using the ANN
model.
ANFIS and ANN models are used to model or predict
some processing variables in the fields of food science and
technology. Although ANFIS and ANN models have been
used for the prediction of some parameters of foods, there are
only a few articles on the comparison of these models. Generally, in these studies, it was determined that the ANFIS or
fuzzy logic models performed better compared to ANN
50
50
ANFIS
Predicted Anthocyanin Values
ANFIS
40
30
20
10
40
30
20
10
10
20
30
40
50
10
30
40
50
40
50
50
50
ANN
ANN
Predicted Anthocyanin Values
20
40
30
20
10
40
30
20
10
10
20
30
40
50
10
20
30
1335
Table 2 RMSE, MAE, and R 2 statistics of the ANFIS and ANN models in training, testing, and validation periods for the estimation of total
anthocyanin contents in black carrot, grape skin, and red cabbage
ANFIS
Data set
Black carrot
Grape skin
Red cabbage
(Number of MFs)
ANN
RMSE
MAE
R2
RMSE
MAE
R2
Training
444 gaussmf
0.0524
0.8325
0.9753
361
1.4428
1.1049
0.9586
Testing
Validation
Training
Testing
Validation
Training
Testing
Validation
444 gaussmf
444 gaussmf
555 gaussmf
555 gaussmf
555 gaussmf
555 gaussmf
555 gaussmf
555 gaussmf
0.0713
0.0746
0.0312
0.0460
0.0467
0.0299
0.0450
0.0457
0.8100
0.8674
0.5167
0.5470
0.5491
0.4905
0.5090
0.5268
0.9787
0.9769
0.9948
0.9944
0.9942
0.9860
0.9842
0.9836
361
361
351
351
351
361
361
361
1.3566
1.3197
1.2943
1.2918
1.2669
0.9967
0.9966
0.9476
1.0366
1.0359
0.9647
0.9469
0.9539
0.7621
0.7471
0.7257
0.9654
0.9674
0.9803
0.9803
0.9809
0.9655
0.9653
0.9683
RMSE root mean square error, MAE mean absolute error, ANFIS adaptive neuro fuzzy inference system, ANN artificial neural network
Conclusion
The results of this study showed that the stability of
anthocyanin-based extracts depended on pH, temperature,
heating time, and anthocyanin sources. Elevated pH levels
cause anthocyanin degradations which affect both color and
intensity (Mazza and Miniati 1993). It was reported that the
thermal stabilities of anthocyanins decreased with the increment of pH (Krca et al. 2007; Torskangerpoll and Andersen
2005) and temperature by several researchers (Kirca et al.
2003; Krca et al. 2007; Wang and Xu 2007). Overall, it can
be stated that an increase in temperature, heating time, and pH
contributed to the loss of anthocyanins. Grape skin anthocyanin content decreased more rapidly than that of black carrot.
The red cabbage anthocyanins exhibited the highest heat and
pH stability. These results indicate that acylated anthocyanins
are more stable at higher temperatures and pH values. There
are several studies on the anthocyanin degradation kinetics of
different anthocyanin sources, such as black carrot (CevallosCasals and Cisneros-Zevallos 2004; Krca et al. 2007), red
cabbage, blackcurrant, elderberry (Cevallos-Casals and
Cisneros-Zevallos 2004), grape (Cevallos-Casals and
Cisneros-Zevallos 2004; Dyrby et al. 2001), and blackberry
juice (Wang and Xu 2007). In this study, in addition to
determining the anthocyanin content and their percent degradation depending on temperature, time, and pH levels, effective predictive models for the estimation of the anthocyanin
content of studied materials were constructed using two different nonlinear modeling techniques (ANFIS and ANN). The
accuracy of the ANFIS and ANN models was compared to
estimate the effects of different temperatures, pH levels, and
time on anthocyanin contents. The results revealed that
ANFIS model is better than the ANN. Both models can be
utilized efficiently for the prediction of unmeasured interval
anthocyanin contents with high accuracy at different temperatures, times, and pH levels.
References
Abu Ghoush M, Samhouri M, Al-Holy M, Herald T (2008) Formulation
and fuzzy modeling of emulsion stability and viscosity of a gum
protein emulsifier in a model mayonnaise system. J Food Eng 84:
348357
Anonymous (2008) Preparing buffer systems. Available from http://
lisanskimya.balikesir.edu.tr/~f10501/tam.htm. Accessed Feb 2008
Arapitsas P, Sjoberg PJR, Turner C (2008) Characterisation of anthocyanins in red cabbage using high resolution liquid chromatography
coupled with photodiode array detection and electrospray
ionization-linear ion trap mass spectrometry. Food Chem 109:219
226
Castaneda-Ovando A, Pacheco-Hernandez M, Paez-Hernandez ME,
Rodriguez JA, Galan-Vidal CA (2009) Chemical studies of anthocyanins: a review. Food Chem 113:859871
Cevallos-Casals BA, Cisneros-Zevallos L (2004) Stability of
anthocyanin-based aqueous extracts of Andean purple corn and
red-fleshed sweet potato compared to synthetic and natural colorants. Food Chem 86:6977
Cigizoglu HK (2003) Estimation, forecasting and extrapolation of flow
data by artificial neural networks. Hydrol Sci J 48:349361
Dai J, Gupte A, Gates L, Mumper RJA (2009) Comprehensive study of
anthocyanin-containing extracts from selected blackberry cultivars:
extraction methods, stability, anticancer properties and mechanisms.
Food Chem Toxicol 47:837847
Dyrby M, Westergaard N, Stapelfeldt H (2001) Light and heat sensitivity
of red cabbage extract in soft drink model systems. Food Chem 72:
431437
Ersus S, Yurdagel U (2007) Microencapsulation of anthocyanin pigments
of black carrot (Daucus carota L.) by spray drier. J Food Eng 80:
805812
1336
Giusti MM, Wrolstad RE (2001) Unit F1.2. In: Wrolstad RE, Schwartz SJ
(eds) Anthocyanins. Characterization and measurement with UV
visible spectroscopy. Current protocols in food analytical chemistry.
Wiley, New York, pp 113
Hagan MT, Menhaj MB (1994) Training feed forward networks with the
Marquardt algorithm. IEEE Trans Neural Networ 6:861867
Haykin S (1998) Neural networksa comprehensive foundation, 2nd
edn. Prentice-Hall, Upper Saddle River, pp 2632
Hernandez JA (2009) Use of neural networks and neural network inverse
in optimizing food processes. CAB Rev: Perspect Agric Vet Sci Nutr
Nat Resour 4(61):111
Hong V, Wrolstad RE (1990) Characterization of anthocyanin-containing
colorants and fruit juices by HPLC/photodiode array detection. J
Agric Food Chem 38:698708
Huang Y, Kangas LJ, Rasco BR (2007) Applications of artificial neural
networks (ANNs) in food science. Crit Rev Food Sci 47:113126
Jang J-SR (1993) ANFIS: adaptive-network-based fuzzy inference system. IEEE Trans Syst Man Cybern 23:665685
Kammerer D, Carle R, Schieber A (2004) Quantification of anthocyanins
in black carrot extracts (Daucus carota ssp. sativus var. atrorubens
Alef.) and evaluation of their color properties. Eur Food Res Technol
219:479486
Karaman S, Ozturk I, Yalcin H, Kayacier A, Sagdic O (2011) Comparison
of adaptive neuro-fuzzy inference system and artificial neural networks for estimation of oxidation parameters of sunflower oil added
with some natural byproduct extracts. J Sci Food Agric 92:4958
Kilic K, Onal-Ulusoy B, Boyaci IH (2007) A novel method for color
determination of edible oils in L*a*b* format. Eur Food Res
Technol 109:157164
Kirca A, zkan M, Cemerolu B (2003) Thermal stability of black carrot
anthocyanins in blood orange juice. J Food Qual 26(5):361366
Krca A, zkan M, Cemeroglu B (2007) Effects of temperature, solid
content and pH on the stability of black carrot anthocyanins. Food
Chem 101:212218
Kisi O (2006) Daily pan evaporation modelling using a neuro-fuzzy
computing technique. J Hydrol 329:636646
Kisi O (2007) Streamflow forecasting using different artificial neural
network algorithms. J Hydrol Eng 12:532539
Maier T, Schieber A, Kammerer DR, Carle R (2009) Residues of grape
(Vitis vinifera L.) seed oil production as a valuable source of
phenolic antioxidants. Food Chem 112:551559
MATLAB (2009) Matlab 7.0.1 for MATLAB reference, neural network
toolbox and fuzzy logic toolbox
Mazza G, Miniati E (1993) Anthocyanins in fruits, vegetables, and grains.
CRC, Boca Raton