THE USE OF K-MEANS CLUSTERING ALGORITHM FOR IDENTIFYING THE TRAFFIC ACCIDENT PATTERNS: CASE OF THE SAKARYA CITY

Yıl 2018, Cilt: 6 Sayı: 3, 89 - 105, 26.12.2018

Samet Güner , Keziban Seçkin Codal , Hüseyin Serdar Geçer , Erman Coşkun

https://doi.org/10.22139/jobs.415578

Cited By: 1

Öz

The most common definition of
traffic accidents is the fatality, injury or damage of one or more vehicles on
the roads (Anderson, 2009). Traffic accidents cause loss of life and property
both in the world and in our country.

A total of 4577 fatal and
injured traffic accidents occurred in the city center between 2006 and 2012 in
Sakarya. A total of 6647 people were injured in these accidents and 62 people
lost their lives (Sakarya Province Traffic Inspection Branch Directorate).
However, it should be noted that this figure only covers citizens who have died
at the site of the accident. This number is expected to increase when the
survivors died in the hospital after the accident.  

In this study, which examines
the traffic accidents occurred in the city of Sakarya between 2006 and 2012,
the similarities between the traffic accidents will be investigated, and the
main accident characteristics will be determined using these similarities.
Therefore, it is aimed to identify similar accident groups and propose specific
solutions to them.

The main objective of the
study is to investigate the similarity relations between traffic accidents
occurring in the city and to provide a road map to the decision makers in the
investments to be made to the developing city infrastructure based on this
similarity relationship. In this study, it is planned to answer the following
research questions:

• How many different clusters
can be collected by using the similarity of Euclidean distance between traffic
accidents?

• What is the weighted
distribution of accidents under these clusters?

• How can the clusters be
classified on the basis of districts?

In this study, K-means
clustering method was used in the analysis of traffic accidents. Clustering
algorithm is a statistical method commonly used in accident analysis. This is
because accidents show many similarities and differences in terms of
environmental characteristics, vehicle characteristics, types of accidents and
driver characteristics. The grouping of these similarities is important to
determine the accident characteristics and to prevent the recurrence of these
accidents. As a matter of fact, it is seen that clustering method is used
widely in the accident analyzes conducted both in the world and in our country.
International studies such as Kim and Yamashita (2007), Anderson (2009),
Bocarejo and Diaz (2011), Figuera et al. (2011) and Mauro et al. (2013) can be
given as example. The studies of Karpat and Yılmaz (2002), Yılmaz and Erişoğlu
(2003), Murat and Şekerler (2009), Atalay and Tortum (2010), Tortum vd. (2011)
and Güner vd. (2014) are the examples of national studies.

K-means clustering method
brings the units together according to similar features, but also removes them
according to their different characteristics. Clustering analysis is a
statistical method for separating units into heterogeneous groups among
themselves by using distance matrix or similarity matrix (Mauro et al. 2013).
In this method, while the distance between the units in the same cluster is
minimized, the distance between the clusters is maximized. Thus, the units
included in the analysis are clustered according to their similar
characteristics, while at the same time they are separated from the other units
due to their different properties. Clustering
algorithm is a statistical method which is commonly used in accident analysis.

The research covers accidents
that occurred between 2006 and 2012. The data set was obtained from Sakarya
Police Department. Since Sakarya Police Department records only fatality or
injured accidents, only property damaged accidents were excluded from the
analysis. In addition, pedestrian accidents were excluded from the analysis and
only vehicle accidents were taken into consideration. Thus, the data set
includes 3333 fatality or injured accidents. 

When the dataset is examined,
it was observed that the most of accidents occurred as side-by-side collision.
Pedestrian accidents also occupy an important place in all accidents. Vehicle
accidents are mostly included the car and heavy car accident. Number of
accidents is similar in summer, spring and autumn, however it decreases in
winter. Majority of these accidents took place under normal circumstances.
According to this result, accidents are occurred in outdoors, during the
daytime and in dry road surface. The majority of accidents occur between 12:00
and 20:00. Due to the high density of
work and the end of working hours, the number of personnel services,
automobiles and public transport vehicles that are on the road cause an
increase in the number of accidents. In addition, municipal crews (such as
cleaning work, road maintenance, water / sewerage works), which work with heavy
maneuvering vehicles during this time of the day, cause traffic density and accidents.

As a result of the clustering
analysis, four different types of accidents were identified for Sakarya in the
context of environmental characteristics, and these types of accidents are
described in detail.

·        
Cluster 1. Collisions involving two vehicles. In this cluster, accidents
took place in autumn, on weekdays and before noon, on dry road surface and
outdoors. Most of accidents in Sakarya is high relevance with Cluster 1.

·        
Cluster 2. Single-vehicle accidents without junction.  These represent the accident that is on wet
road surface, a cloudy day in autumn, on weekdays and midnight. The cluster is
unique for mapping of single-vehicle accident.

·        
Cluster 3. Two-vehicle accidents with junction_a.  The features of accident in Cluster 3 are on
dry road surface, outdoors, in summer, at weekends and afternoon. The traffic
density in Cluster 3 is more than other type of Clusters.

·        
Cluster 4. Two-vehicle accidents with junction_b. In Cluster 4,
the accidents are occurred dry road surface, outdoors, in spring, on weekdays
and afternoon.

The percentages of Clusters
are 46%, 10%, 27% and %17, respectively. The accident density of Cluster 1 is
high when compared with the others. The most attractive features of Cluster 1
are on weekdays and before noon and Cluster 3 that has second density of
accident of all is on junction, in summer, at weekend and afternoon.

Hence, the results of the
research revealed the basic characteristics of the traffic accidents that took
place in Sakarya.  The accidents in
Adapazarı, Erenler, Serdivan ve Arifiye that is sub-province match up with the
clusters to determine the characteristics of accident. Sub-provinces correspond
with Clusters such as Adapazarı-Cluster 1, Serdivan-Cluster 3, Erenler-Cluster
4 and Arifiye- Cluster 2. Thus, it is expected that this research will help
decision makers to prevent and reduce traffic accidents.

TRAFİK KAZA DESENLERİNİN TANIMLANMASINDA K-MEANS KÜMELEME ALGORİTMASININ KULLANILMASI: SAKARYA İLİ UYGULAMASI

Öz

Amaç: Trafik
kazaları hem dünyada hem de ülkemizde can ve mal kayıplarına neden olmaktadır.
Sakarya ilinde meydana gelen trafik kazalarının incelendiği bu çalışmada,
şehirde meydana gelen trafik kazaları arasındaki benzerlik ilişkileri
araştırılmış ve bu benzerlik ilişkisinden yola çıkarak belli başlı kaza
karakteristikleri belirlenmiştir. Böylelikle, benzer kaza gruplarının
belirlenmesi ve bunlara özgü çözüm önerileri getirilmesi amaçlanmaktadır.  

Yöntem: Trafik
kazalarının analizinde K-means kümeleme yöntemi kullanılmıştır. Bu yöntem,
birimleri benzer özelliklerine göre bir araya getirirken, farklı özelliklerine
göre de uzaklaştırmaktadır. 

Bulgular: Kümeleme
analizi sonucunda, çevresel özellikler bağlamında Sakarya iline özgü dört
farklı kaza türü olduğu tespit edilmiş ve bu kaza türleri detaylı olarak
tanımlanmıştır.  Sakarya genelinde en
tipik kazaların kavşağın olmadığı yerlerde, kuru zeminde, açık havada, sonbahar
mevsiminde, hafta içi ve öğleden önce gerçekleştiği belirlenmiştir. Ayrıca,
analize dâhil olan Adapazarı, Erenler, Serdivan ve Arifiye ilçeleri için de
tipik kaza karakteristikleri tanımlanmıştır.

Sonuç: Araştırma
sonuçları, Sakarya’da meydana gelen trafik kazalarının temel
karakteristiklerini ortaya koymuştur. Böylelikle, araştırmanın trafik kazalarının
önlenmesinde ve azaltılmasında karar vericilere yardımcı olması beklenmektedir. 

Anahtar Kelimeler

Kümeleme, K-means, Trafik, Kaza

Kaynakça

• Atalay, A. ve Tortum, A. (2010). Türkiye’deki İllerin 1997-2006 Yılları Arası Trafik Kazalarına Göre Kümeleme Analizi. Pamukkale Üniversitesi Mühendislik Bilimleri Dergisi, 16(3): 335-343.
• Anderson, T.K. (2009). Kernel Density Estimation and K-Means Clustering to Profile Road Accident Hotspots. Accident Analysis and Prevention, 41: 359-364.Bocajero, J.P. ve Diaz, C.A. (2011). Characterization of Fatal Road Traffic Accidents Using K-Means Clustering – A Case Study of Bogotá. TRB Annual Meeting, Washington D.C.
• Figuera, C., Lillo, J.M., Mora-Jiménez, I., Rojo-Álvarez, J.L. ve Caamaño, A.J. (2011). Multivariate Spatial Clustering of Traffic Accidents for Local Profiling of Risk Factors. 14th International IEEE Conference on Intelligent Transportation Systems, Washington D.C.
• Güner, S., Geçer, H.S. ve Coşkun, E. (2014). Toplu Taşıma Araçlarının Dâhil Olduğu Trafik Kazalarının K-Means Kümeleme Algoritması ile Analizi: Sakarya Uygulaması. TRANSIST 7. Uluslararası Ulaşım Teknolojileri Sempozyum ve Fuarı, İstanbul.
• Karpat, G. ve Yılmaz, V. (2002). Türkiye'deki Trafik Kazaları Oluş Şekillerinin, Kazanın Olduğu Yerdeki Trafik, Aydınlatma ve Yol Durumuna Göre İller Bazında İncelenmesi. Uluslararası Trafik ve Yol Güvenliği Kongresi, Ankara.
• Kim, K. ve Yamashita, E.Y. (2007). Using a K-Means Algorithm to Examine Patterns of Pedestrian Involved Crashes in Honolulu. Hawaii, Journal of Advanced Transportation, 41(1): 69-89.
• Li, T., Chen, Y., Qin, S. ve Li, N. (2011). Highway Road Accident Analysis Based on Clustering Ensemble. İçinde Y. Yu, Z. Yu, ve J. Zhao (Ed.), Communications in Computer and Information Science, Springer.
• Mauro, R., Luca, M.D. ve Dell’Acqua, G. (2013). Using a K-Means Clustering Algorithm to Examine Patterns of Vehicle Crashes in Before-After Analysis. Modern Applied Science, 7(10): 11-19.
• Murat, Y.Ş. ve Şekerler, A. (2009). Trafik Kaza Verilerinin Kümelenme Analizi Yöntemi ile Modellenmesi, İMO Teknik Dergi, Yazı 311: 4759-4777.
• Tan, P-N., Steinbach M. ve Kumar V. (2006). Introduction to Data Mining, Pearson Education Limited.
• Tortum, A., Kabakuş, N., Codur, M.Y., Atalay, A. ve Uluğtekin, N. (2011). Clustering Analysis of the Districts in Erzurum for Traffic Accidents between 2002 and 2007. Scientific Research and Essays, 6(13): 2850-2857.
• World Health Organization, Global Status Report on Road Safety (2015). Switzerland.
• Yılmaz, V. ve Erişoğlu, M. (2003). Türkiye’de Trafik Kazalarında Riskli İllerin İstatistiksel Olarak Belirlenmesi. Afyon Kocatepe Üniversitesi Fen Bilimleri Dergisi, 3(1-2): 129-146.