ORIGINAL ARTICLE

Automatic determination of traffic accidents based on KMC-basedattribute weighting

Kemal Polat • S. Savas Durduran

Received: 5 September 2010 / Accepted: 7 February 2011 / Published online: 24 February 2011

� Springer-Verlag London Limited 2011

Abstract In this study, the traffic accidents recognizing

risk factors related to the environmental (climatological)

conditions that are associated with motor vehicles acci-

dents on the Konya-Afyonkarahisar highway with the aid

of Geographical Information Systems (GIS) have been

determined using the combination of K-means clustering

(KMC)-based attribute weighting (KMCAW) and classifier

algorithms including artificial neural network (ANN) and

adaptive network-based fuzzy inference system (ANFIS).

The dynamic segmentation process in ArcGIS9.0 from the

traffic accident reports recorded by District Traffic Agency

has identified the locations of the motor vehicle accidents.

The attributes obtained from this system are day, temper-

ature, humidity, weather conditions, and month of occurred

traffic accidents. The traffic accident dataset comprises five

attributes (day, temperature, humidity, weather conditions,

and month of occurred traffic accidents) and 358 obser-

vations including 179 without accident and 179 with

accident. The proposed comprises two stages. In the first

stage, the all attributes of dataset have been weighted using

KMCAW method. The aims of this weighting method are

both to increase the classification performance of used

classifier algorithm and to transform from linearly non-

separable traffic accidents dataset to a linearly separable

dataset. In the second stage, after weighting process, ANN

and ANFIS classifier algorithms have been separately used

to determine the case of traffic accidents as with accident

or without accident. In order to evaluate the performance of

proposed method, the classification accuracy, sensitivity,

specificity and area under the ROC (Receiver Operating

Characteristic) curves (AUC) values have been used. While

ANN and ANFIS classifiers obtained the overall prediction

accuracies of 53.93 and 38.76%, respectively, the combi-

nation of KMCAW and ANN and the combination of

KMCAW and ANFIS achieved the overall prediction

accuracies of 74.15 and 55.06% on the prediction of traffic

accidents. The experimental results have demonstrated that

the proposed attribute weighting method called KMCAW

is a robust and effective data pre-processing method in the

prediction of traffic accidents on Konya-Afyonkarahisar

highway in Turkey.

Keywords Geographical information systems (GIS) �Traffic accident analysis � Prediction � K-means clustering

based attribute weighting � Artificial neural network �Adaptive network based fuzzy inference system

1 Introduction

Today, the most negative result of developing transporta-

tion systems is traffic accidents with injuries and loss of

lives. The tremendous social and economic costs associated

with traffic accidents have led many road authorities and

researchers to establish safety management programs that

aim to continually improve the safety performance of

highways [1]. So, traffic safety is the most critical issue in

agencies’ transportation strategy. The identification of

safety deficient areas on the highways is aimed to

K. Polat (&)

Engineering of Faculty, Electrical and Electronics Engineering

Department, Bartin University, 74100 Bartin, Turkey

e-mail: kemal_polat2003@yahoo.com

S. S. Durduran

Department of Geomatics Engineering, Selcuk University,

42035 Konya, Turkey

e-mail: durduran@selcuk.edu.tr

123

Neural Comput & Applic (2012) 21:1271–1279

DOI 10.1007/s00521-011-0559-9

implement precautionary measures and provisions by

researchers and traffic officials [2].

Geographical Information System (GIS) technology is

becoming an increasingly popular tool for visualization and

analyses of accident data in highways. GIS has the ability

to hold a vast amount of data that can be easily stored,

shared, analyzed, and managed. It provides a platform for

spatial data analyses and visualization to explore relation-

ships between spatial and no spatial data [2].

The success of safety improvement programs in

reducing accident occurrence depends on the methods

used in the accident analyses. Today, many researchers

used different deterministic and statistical methods in the

studies that aiming to determine the high rate accident

locations and safety deficient areas on the highways

[2–4]. Many of these researches have explored the rela-

tionships between traffic accidents and geometric design

and operation of road segments. However, data relating to

accidents are widely available, but have received sur-

prisingly little analyses with respect to weather. The

complexity involved in establishing the exact cause-and-

effect relationship in traffic accidents acts as an obstacle,

because road accidents are the results of an intricate

driver–vehicle–environment matrix [5]. So, it is aimed to

decrease the accidents determining the effects of weather

and some environmental conditions on traffic accidents in

Konya-Afyonkarahisar highway with the aid of GIS and

artificial intelligence.

There are a few papers related to classification of traffic

accidents in literature. Among these, Durduran et al. have

proposed a decision making system (DMS) based on cor-

relation-based feature selection and classifier algorithms

including support vector machine (SVM) and artificial

neural network (ANN) has been proposed to predict the

traffic accidents in Konya-Afyonkarahisar highway. They

obtained the prediction accuracy of 61.79% with ANN

classifier and achieved the prediction accuracy of 67.42%

using SVM with RBF (radial basis function) kernel [20].

Polat et al. have used the classifier ensemble methods

including SVM (Support Vector Machine) ensemble, ANN

ensemble, and ANFIS ensemble methods. In the classifi-

cation of the traffic accidents, they obtained 70.22, 91.01,

91.01% success rates using geometrical mean, arithmetical

mean, and majority voting methods with SVM ensemble,

separately while they achieved 73.03, 82.55, 82.55% suc-

cess rates with ANN ensemble method. As for ANFIS

ensemble, they obtained the success rates of 66.85, 80.33,

and 80.89% using geometrical mean, arithmetical mean,

and majority voting methods, separately [21].

In this article, a novel hybrid system based on KMCAW

and classifier algorithms including ANN and ANFIS has

been proposed and applied to determination of traffic

accidents using GIS. The aims of this study are to increase

the discrimination ability of between without traffic acci-

dents and with traffic accidents using KMCAW prior to

classifier algorithm and to transform from linearly non-

separable dataset to linearly separable dataset. In training

and testing of ANN and ANFIS classifiers, the 50–50%

split of whole dataset has been randomly chosen. In order

to test the performance of proposed method, the classifi-

cation accuracy, sensitivity, specificity, and AUC (area the

ROC curve) values have been used. Two hybrid methods

called the combination of KMCAW–ANN and the com-

bination of KMCAW–ANFIS have been used and chosen

the combination of KMCAW–ANN as the best method in

determination of traffic accidents belonging to Konya-Af-

yonkarahisar highway.

The remaining of the paper is organized as follows. The

material is presented in the next section. The method is

explained in Sect. 3. The experimental data and results to

present the effectiveness of proposed method are given in

Sect. 4. Finally, the conclusions are given in Sect. 5 with

future directions.

2 Traffic accident database and used material

2.1 Traffic accidents dataset

Afyonkarahisar-Konya highway is a junction region in

Turkey connecting the industrial, tourism and agricultural

areas to each other. Especially in winter, weather-related

crashes happen frequently because of continental climate in

the region. The length of the highway is approximately

240 km. 65% of the road has two lanes. Traffic accident

reports of the highway are obtained on paper form by

District Traffic Agency Officers in Turkey. These records

include collected accident parameters such as the date,

hour/minutes, kilometer of crash, code of highway, age,

sex and alcohol consumption of driver, weather conditions,

lighting conditions, vehicle type, number of persons

injured/killed, etc. Figure 1 presents the Konya-Afyonka-

rahisar highway in Turkey. Table 1 presents the statistical

values of attributes of raw traffic accident dataset [20].

Table 2 shows the statistical values of attributes of

weighted traffic accident dataset by K-means clustering-

based attribute weighting.

A GIS-based study with accident data has been con-

ducted with the aim of reducing the number of the acci-

dents by determining the effects of environment and

weather conditions on the phenomenon of accidents. First,

179 accidents and 179 not accident records of the year

2006 were collected from the District Traffic Agency and

stored into an MS Access database. Meanwhile, highway

was digitized at a scale of 1:1,000,000 with Arc GIS 9.0

software. The location of the accidents is positioned with

1272 Neural Comput & Applic (2012) 21:1271–1279

123

the ‘‘kilometer of crash’’ data on the route of highways

using the ‘‘linear referencing’’ toolset in Arc GIS 9.0 [20].

3 Method

In this paper, we have proposed a hybrid method com-

prising attribute weighting (KMCAW) and classification

stage (ANN and ANFIS classifier algorithms) to determine

the traffic accidents on Konya-Afyonkarahisar highway in

Turkey. Figure 2 demonstrates the block diagram of pro-

posed method.

3.1 Weighting stage: K-means clustering-based

attribute weighting

In K-means clustering-based attribute weighting method, at

first the clusters of each attribute are found using K-means

clustering (KMC) and calculated the distance between its

cluster and mean value of that attribute. According to

calculated distances, attributes are weighted [19].

The goal of attribute weighting method is to map the

attributes according to their distributions in a dataset and

also transform from linearly non-separable dataset to lin-

early separable dataset. Attribute weighting method works

based upon principle that decreasing the variance in attri-

butes forming dataset. Thanks to this weighting method,

Fig. 1 Konya-Afyonkarahisar

highway in Turkey [20, 21]

Table 1 The statistical values of attributes of raw traffic accident

dataset [20]

Attribute number Minimum Maximum Mean Standard deviation

1 1 7 4.23 1.97

2 -4 29 14.36 8.31

3 20 91 53.53 21.74

4 1 5 2.502 1.29

5 1 12 7.74 3.16

N = 358 observations comprising 179 without accident and 179 with

accident

Table 2 The statistical values of attributes of weighted traffic acci-

dent dataset by KMCAW

Attribute number Minimum Maximum Mean Standard deviation

1 0.89 7.72 4.27 2.08

2 -7.23 52.42 17.98 13.83

3 14.28 139.72 60.89 34.95

4 0.836 6.12 2.566 1.398

5 0.96 12.45 7.75 3.17

N = 358 observations comprising 179 without accident and 179 with

accident

Traffic accident dataset with two classes including “without accidents”

and “with accidents”

Weighting of traffic accident dataset using KMCAW method

Determination of weighted traffic accidents using ANN and ANFIS

Without accident or

With accident

Results:

Fig. 2 The flow chart of the proposed method

Neural Comput & Applic (2012) 21:1271–1279 1273

123

the similar data in same attribute are gathered and the

discrimination ability of classifier is increased [19].

K-means clustering also known as C-means clustering

has been applied to a variety of areas including image

segmentation, speech data compression, data mining, etc.

[6, 7, 8, 9, 10]. The working of KMC can be summarized as

follows [11, 19]:

Phase 1 Choose K initial cluster centers z1, z2,…,zK

randomly from the n points{X1, X2, X3,…,Xn}

Phase 2 Assign point Xi, i = 1, 2,…,n to the cluster Cj,

j [ {1, 2,…,K}

if Xi � zj

��

��\ Xi � zp

��

��; p ¼ 1; 2; . . .;K

and j 6¼ p

Phase 3 Compute new cluster centers as follows

znewi ¼ 1

ni

P

Xj2Ci

Xj i ¼ 1; 2; . . .;K

where ni is the number of elements belonging to

the cluster Ci

Phase 4 If znewi � zi

��

��\e; i ¼ 1; 2; . . .;K, then

terminate. Otherwise continue from phase 2

This weighting method works as follows: first, the

cluster centers are calculated using KMC method. After

computing the centers of attributes, the ratios of means of

attributes to their centers are calculated and these ratios are

multiplied with data point of each attribute. The readers

can refer to [19] for more information about this weighting

method. Figure 3 demonstrates the flowchart of KMCAW

method.

3.2 Classification stage

After KMCAW process, the classifier algorithms including

ANN and ANFIS classifiers are used. These algorithms

have been explained in the following subsections. In the

training and testing of classifiers, 50–50% training–testing

split has been used.

3.2.1 Artificial neural network (ANN) with Levenberg

Marquart (LM)

An ANN is constructed for a specific application, such as

pattern recognition or data mining, by means of a learning

process. The back propagation (BP) algorithm is a most

widely used training procedure that adjusts the connection

weights of a multi-layer perceptron (MLP) [12]. Mainly,

the LM algorithm is a least-squares estimation algorithm

on the basis of the maximum neighborhood idea. An MLP

comprises three layers: an input layer, an output layer, and

one or more hidden layers. Each layer comprises a deter-

mined number of neurons. The neurons in the input layer

only work as buffers for distributing the input signals xi to

neurons in the hidden layer [12].

In applications, the number of node in input layers and

the number of node in output layers change according to

the number of samples and class labels in used dataset,

respectively. The number of node in hidden layer has been

determined by trial and error method. Table 3 presents the

number of node in the input, hidden, and output layers on

network in prediction of traffic accident dataset using

MLPANN.

3.2.2 Adaptive network-based fuzzy inference system

(ANFIS)

ANFIS are a class of adaptive networks that are function-

ally equivalent to fuzzy inference systems. ANFIS is an

adaptive version of fuzzy inference system based on

combining neural network learning abilities and inference

ability of fuzzy logic. Especially, ANFIS structure consists

of representing as a network that has neural learning ability

of Sugeno type fuzzy systems. Figure 4 presents a basic

Sugeno ANFIS structure comprising five layers.

Loading Datasets ----------------------------------------------------------------------- Load the dataset including classes

Obtaining Centers of All attributes --------------------------------------------------------------------- Find the centers of attributes using K-means Clustering (KMC)

Weighting Process ------------------------------------------------------------------------

Calculate the ratios of means of attributes to their centers and multiply these ratios with each attribute

Finding final weighting ratios -------------------------------------------------------------------------Weighting of attributes using KMC method

Fig. 3 The block diagram of KMC. The flowchart of K-means

clustering (KMC)-based attribute weighting

Table 3 The number of nodes in the input layer, hidden layer, and

output layer on MLP network

Dataset The number of

node in input

layer

The number of

node in hidden

layer

The number of

node in output

layer

Traffic

accident

dataset

5 10 2

1274 Neural Comput & Applic (2012) 21:1271–1279

123

In training of ANFIS learning algorithm, to generate the

FIS structure, the subtractive clustering algorithm has been

used. Optimization method used in membership function

parameter training (FIS parameters) was the hybrid

method, which is a combination of least-squares estimation

with backpropagation. In this way, the best ANFIS model

has been achieved and obtained good results in classifica-

tion of traffic accidents. For more information on this

structure, the readers are referred to [13–15, 23].

4 Results and discussion

4.1 Benchmark dataset and performance measures

In this section, a novel attribute weighting called KMCAW

has been proposed to classify the traffic accidents dataset as

with accident and without accident. Also, in addition to

traffic accidents dataset, Iris dataset taken from UCI

machine learning database [22] has been used as bench-

mark dataset to further evaluate the propose method. In

order to test the performance of proposed method, the

classification accuracy, sensitivity, specificity and AUC

values have been used.

4.1.1 Benchmark dataset

Iris dataset: Iris dataset is the best known database to be

found in the pattern recognition applications. The dataset

contains 3 classes of 50 instances each, where each class

refers to a type of Iris plant. In Iris dataset [22]:

Number of attributes: 4

Number of instances: 150

Number of classes: 3

4.1.2 Performance measures

As the performance measures, the classification accuracy,

sensitivity, specificity and AUC values have been used and

explained as follows. In training and testing of ANN and

ANFIS classifiers, 50–50% train–test data split has been

used and randomly chosen from dataset. The confusion

matrix is shown in Table 4 (Actual vs. Predicted) and the

other parameters which are computed using confusion

matrix are shown with the following formulas, where TP,

TN, FP and FN denote true positives, true negatives, false

positives, and false negatives, respectively [16, 17].

For classification accuracy, sensitivity and specificity

analysis, and f-measure, the following expressions have

been used.

Classification Accuracy ð%Þ ¼ TPþ TN

TPþ FPþ TNþ FN� 100

ð1Þ

Sensitivity ð%Þ ¼ TP

TPþ FN� 100 ð2Þ

Specificity ¼ TN

FPþ TN� 100 ð3Þ

ROC (Receiver Operating Characteristic) curves and

AUC value: A receiver operating characteristic (ROC)

graph is a technique for visualizing, organizing and

selecting classifiers based on their performance. ROC

graphs are commonly used in medical decision making,

and have been used increasingly in machine learning and

data mining research in recent years [18]. ROC graphs are

two-dimensional graphs in which tp (true positive) rate is

plotted on the Y axis and fp (false positive) rate is plotted

on the X axis. An ROC depicts relative tradeoffs between

benefits (true positives) and costs (false positives) [18]. In

ROC curves, the performance criteria is AUC (under the

ROC curve) value. The higher the AUC value, the higher

will be the classification performance.

4.2 Results

In this paper, a novel hybrid method based on combining of

means clustering-based attribute weighting and classifier

algorithms (ANN and ANFIS classifiers) has been pro-

posed and applied to determination of motor vehicle

accidents on the Konya-Afyonkarahisar highway with the

aid of Geographical Information Systems. In order to

increase the classification performance and to transform

from non-linear traffic accidents dataset to a linearly sep-

arable dataset, the attribute weighting method called

KMCAW has been used prior to classification stage.

Through this weighting method, the variance value

belonging to attributes within each class has been

decreased. In this way, the data within each attribute have

Fig. 4 A basic Sugeno ANFIS structure comprising five layers [14]

Table 4 Confusion matrix

Confusion matrix No Yes

No True negative (TN) False positive (FP)

Yes False negative (FN) True positive (TP)

Neural Comput & Applic (2012) 21:1271–1279 1275

123

been gathered and then the classification performance is

increased. In working of KMCAW, the centers of attributes

in dataset have been found with KMC. And then, the ratios

of means of attributes to their centers have been calculated

and these ratios have been multiplied by each attribute in

dataset. Figure 5 demonstrates the distribution of raw

traffic accident dataset and weighted traffic accident dataset

according to first three attributes (1st, 2nd, and 3rd). Fig-

ure 6 depicts the box graph presentations of raw and

weighted traffic accidents dataset with 5 attributes. In order

to show the superiority of the proposed method, the prin-

cipal component analysis (PCA) has been used. PCA has

been applied to both raw and weighted traffic accidents

dataset. The principal components have been ordered (from

higher value to lower value) according to their variance

values of PCs (principal components). In this way, the first

three components have been processed and plotted as 3D

showing the class distribution of traffic accidents dataset.

Figure 7 demonstrates the component distribution of raw

traffic accidents dataset using PCA. Figure 8 exhibits

component distribution of weighted traffic accidents data-

set using PCA. Table 5 shows the variance values of raw

and weighted traffic accidents dataset for first three prin-

cipal components using PCA. For data visualization of

traffic accidents dataset, the PCA has been used. As can be

seen from Fig. 5, the proposed weighting method has

abilities of effective and distinctive attributes in deter-

mining of traffic accidents.

Afterwards, the weighted traffic accident dataset has

been applied to classifier algorithms including ANN and

ANFIS classifiers. In the training and testing process of

classifiers, the 50–50% training–testing split has been used.

Table 6 shows the obtained results from ANN, ANFIS, the

combination of KMCAW and ANN, and combination of

KMCAW and ANFIS using 50–50% training–test partition

on the classification of traffic accidents in a GIS platform.

Table 7 displays the confusion matrixes for ANN, ANFIS,

Fig. 6 a The box graphpresentations of raw and

weighted traffic accidents

dataset with 5 attributes. b The

box graph presentation of

weighted traffic accidents

dataset with KMCAW

Fig. 5 a The distribution of raw traffic accident dataset according to

first three attributes (1st, 2nd, and 3rd attributes). b The distribution of

weighted traffic accident dataset with KMCAW method according to

first three attributes (1st, 2nd, and 3rd attributes)

1276 Neural Comput & Applic (2012) 21:1271–1279

123

Fig. 8 The component

distribution of weighted traffic

accidents dataset using PCA

Fig. 7 The component

distribution of raw traffic

accidents dataset using PCA

Table 6 The obtained results from ANN, ANFIS, the combination of KMCAW and ANN, and the combination of KMCAW and ANFIS using

50–50% training-test partition on the classification of traffic accidents in a GIS platform

Method Classification accuracy

(%)

Sensitivity

(%)

Specificity

(%)

Area under the ROC

curve

Artificial neural network (ANN) 53.93 52.67 57.44 0.539

Adaptive network based fuzzy Inference system

(ANFIS)

38.76 26.19 42.64 0.386

Combination of KMCAW and ANN 74.15 65.92 100 0.742

Combination of KMCAW and ANFIS 55.06 56.16 54.28 0.551

Table 5 The variance values of raw and weighted traffic accidents dataset for first three principal components using PCA

Used dataset Principal components PC1 PC2 PC3 PC4 PC5

Raw traffic accident dataset Variance value 2.137 1.171 0.948 0.58 0.159

Weighted traffic accident dataset with KMC method Variance value 2.313 1.074 0.8424 0.5269 0.242

Neural Comput & Applic (2012) 21:1271–1279 1277

123

the combination of KMCAW and ANN, and combination of

KMCAW and ANFIS with 50–50% training-test partition

on the classification of traffic accidents in a GIS platform.

In addition to traffic accidents dataset, we have used Iris

dataset taken from UCI machine learning database [22] as

benchmark dataset to evaluate the proposed method. As

classifier algorithm, k-NN (k-nearest neighbor) classifier

algorithm has been used after weighting process. The

obtained results are presented in Table 8.

While the data distribution of raw traffic accident

dataset has a linearly non-separable distribution, the data

distribution of weighted traffic accident dataset with

KMCAW is close to a linearly separable distribution. Also,

the obtained results have shown that the proposed

weighting method could be confidently used prior to clas-

sification algorithm in the classification of traffic accidents.

5 Conclusions

Due to the loss of lives and large amount of money, the

researches intending to prevent the traffic accidents are

very popular in developing countries. In this situation, it is

very important to predict the probability of the occurrence

of accidents according to the weather conditions for

implementing proper precautionary measures in traffic

safety studies.

In this study, we have proposed two-stage classification

algorithm based on attribute weighting (KMCAW) and

classifier algorithm (ANN and ANFIS classifiers) to

determine the traffic accidents with aid of GIS platform on

the Konya-Afyonkarahisar highway in Turkey. In the first

stage, the all attributes of dataset have been weighted using

KMCAW method both to increase the classification per-

formance of used classifier algorithm and to transform

from linearly non-separable traffic accidents dataset to a

linearly separable dataset. In the second stage, after

weighting process, ANN and ANFIS classifier algorithms

have been separately used to determine the case of traffic

accidents as with accident or without accident. In order to

ensure the superiority of proposed method, data visuali-

zation methods including box graph representations and

PCA (principal component analysis) have been used. The

proposed method has obtained very good results and shown

the applicability to determination of traffic accidents. So, it

is aimed to obtain more reliable results for the system with

collecting more accident data, according to the weather and

environment conditions as future work.

References

1. Sawalha Z, Sayed D (2006) Transferability of accident prediction

models. Saf Sci 44:209–219

2. Erdogan S, Yilmaz I, Baybura T, Gullu M (2008) Geographical

information systems aided traffic accident analysis system case

study: city of Afyonkarahisar. Accid Anal Prev 40(1):174–181

3. Levine N, Kim KE, Lawrence HN (1995) Spatial analysis of

Honolulu motor vehicle crashes: I. Spatial patterns. Accid Anal

Prev 27(5):663–674

4. Loo BPY (2006) Validating crash locations for quantitative

spatial analysis: a GIS based approach. Accid Anal Prev

38(1):879–886

Table 7 Confusion matrixes for ANN, ANFIS, the combination of KMCAW and ANN, and combination of KMCAW and ANFIS with 50–50%

training-test partition on the classification of traffic accidents in a GIS platform

Output\desired Result (without accident) Result (with accident) Classifier methods

Result (without accident) 27 62 ANN classifier

Result (with accident) 20 69

Result (without accident) 58 31 ANFIS classifier

Result (with accident) 78 11

Result (without accident) 43 46 Combination of KMCAW and ANN

Result (with accident) 0 89

Result (without accident) 57 32 Combination of KMCAW and ANFIS

Result (with accident) 48 41

Table 8 The obtained results in classification of raw Iris dataset and

weighted Iris dataset using k-NN classifier

Used dataset Raw Iris dataset

(without preprocessing)

Weighted Iris dataset

with KMC algorithm

Performance

measure

Classification accuracy

(%)

Classification accuracy

(%)

k = 1 in k-NN

classifier

94.67 100

k = 2 in k-NN

classifier

92.00 100

k = 3 in k-NN

classifier

92.00 100

k = 4 in k-NN

classifier

94.67 100

k = 5 in k-NN

classifier

92.00 100

1278 Neural Comput & Applic (2012) 21:1271–1279

123

5. Andreescu MP, Frost DB (1998) Weather and traffic accidents in

Montreal, Canada. Clim Res 9:225–230

6. MacQueen B (1967) Some methods for classification and analysis

of multivariate observations. Proceedings of 5th Berkeley sym-

posium on mathematical statistics and probability, University of

California Press, Berkeley, 1:281–297

7. Bezdek JC (1981) Pattern recognition with fuzzy objective

function algorithms. Plenum Press, New York

8. Yager RR, Filev DP (1994) Generation of fuzzy rules by

mountain clustering. IEEE Trans Syst Man Cybern 24:209–219

9. Chiu SL (1994) Fuzzy model identification based on cluster

estimation, J Intell Fuzzy Syst 2

10. Rui Xu, Donald Wunsch II (2005) Survey of clustering algo-

rithms. IEEE Trans Neural Netw 16(3):645–678

11. Guldemır H, Sengur A (2006) Comparison of clustering algo-

rithms for analog modulation classification. Expert Syst Appl

30(4):642–649

12. Dirgenali F, Kara S (2006) Recognition of early phase of ath-

erosclerosis using principles component analysis and artificial

neural networks from carotid artery Doppler signals. Expert Syst

Appl 31(3):643–651

13. Jang JSR (1992) Self-learning fuzzy controllers based on

temporal backpropagation. IEEE Trans Neural Netw 3(5):

714–723

14. Jang JSR (1993) ANFIS: adaptive-network based fuzzy inference

systems. IEEE Trans Syst Man Cybern 23(03):665–685

15. Tsoukalas LH, Uhrıg RE (1996) Fuzzy and neural approaches in

engineering. Wiley, New York

16. Lewis D, Gale W (1994) A sequential algorithm for training text

classifiers. Annual ACM conference on research and develop-

ment in information retrieval, the 17th annual international ACM

SIGIR conference on research and development in information

retrieval, New York, 3–12

17. Ma Y, Guo L, Cukic B (2006) A statistical framework for the

prediction of fault-proneness, advances in machine learning

application in software engineering. Idea Group Inc

18. Fawcett F (2005) An introduction to ROC analysis. Pattern Re-

cogn Lett 27(8):861–874

19. Gunes S, Polat K, Yosunkaya S (2010) Efficient sleep stage

recognition system based on EEG signal using k-means clustering

based feature weighting. Expert Syst Appl 37(12):7922–7928

20. Durduran SS (2010) A decision making system to automatic

recognize of traffic accidents on the basis of a GIS platform.

Expert Syst Appl 37(12):7729–7736

21. Polat K, Durduran SS (2010) Ensemble of classifiers for intelli-

gent recognition of traffic accidents using a geographical infor-

mation systems platform, 1st international symposium on

computing in science and Engineering (ISCSE), June 3–5, 2010,

Kusadası, Turkiye

22. Frank A, Asuncion A (2011) UCI machine learning repository.

University of California, School of Information and Computer

Science, Irvine, CA. http://archive.ics.uci.edu/ml

23. Polat K, Gunes S (2006) A hybrid medical decision making

system based on principles component analysis, k-NN based

weighted pre-processing and adaptive neuro-fuzzy inference

system. Digital Signal Process 16(6):913–921

Neural Comput & Applic (2012) 21:1271–1279 1279

123