Please use this identifier to cite or link to this item: https://hdl.handle.net/20.500.13087/825
Title: Prediction of fatal traffic accidents using one-class SVMs: a case study in Eskisehir, Turkey
Authors: Erzurum Çiçek, Zeynep İdil
Kamışlı Öztürk, Zehra
Keywords: Fatal traffic accidents
prediction
one-class classification
binary classification
variable selection
Issue Date: 2021
Publisher: Taylor & Francis Ltd
Abstract: The objective of this study is to investigate the applicability of one-class classification (OCC) models in traffic accident prediction. So far, the accident prediction problem has been considered as a binary classification problem in the literature. Since real accident datasets often involve only accident situations, we thought that OCC could provide more successful predictions. In this study, the fatal accidents, which occurred in Eskisehir, Turkey between 2005 and 2012 was considered. The accidents were tried to be predicted using one-class Support Vector Machine (SVM). In order to compare the performance of the OCC model, some most used binary classifiers were used. Additionally, a non-accident generation procedure was defined to add non-accident cases to the accident dataset. After training, tests were performed using one-class and binary classifiers for the test set generated from the extended dataset. As a result, the one-class SVM model outperformed the binary classification models. Besides, true and false accident alarms were also calculated. The alarm rates obtained with the OCC model also demonstrated that OCC can be suitable for accident prediction rather than binary classification.
URI: https://doi.org/10.1080/13588265.2021.1959168
https://hdl.handle.net/20.500.13087/825
ISSN: 1358-8265
1754-2111
Appears in Collections:İstatistik Bölümü Koleksiyonu
Scopus İndeksli Yayınlar Koleksiyonu
WoS İndeksli Yayınlar Koleksiyonu

Show full item record

CORE Recommender

Page view(s)

16
checked on Oct 3, 2022

Google ScholarTM

Check

Altmetric


Items in GCRIS Repository are protected by copyright, with all rights reserved, unless otherwise indicated.