Geographic information system-based analysis of factors affecting pedestrian-vehicle accidents in urban road segments: Adana case

In this study, to improve the pedestrian safety, it was aimed to determine the factors that affect the occurrence of pedestrian-vehicle accidents on urban road segments and to evaluate the model performances of different machine learning techniques in risk analysis performed on critical road segments which were determined with the help of Geographic Information Systems. For this purpose, the traffic accident data of Adana province taken from the General Directorate of Police in 2013-2017 were analyzed in detail and the single vehicle-single pedestrian accidents that occurred on the road segments were separated according to peak hours. Single vehicle-single pedestrian accidents were mapped in the Geographic Information Systems for 4 peak hour types and Planar Kernel Density Analysis was performed. In the road segments that are critically identified as a result of the analyses; data including 13 different parameters on road geometry and road characteristics, traffic characteristics, the built environment, and land use were obtained through on-site observations and measurements. The effect of the parameters on pedestrian safety was evaluated with the C4.5 Decision Tree model, Multilayer Perceptron and Converted Fuzzy-Decision model. As a result of the study, it was determined that the percentage of public transport, road type, speed, land use, parking and peak hour type parameters have a high impact in terms of pedestrian safety in urban road segments. When the performances of the models were compared, it was seen that the hybrid Converted Fuzzy- Decision model showed the highest performance with 61.42% accuracy and 58.31% weighted average F1-Score. In light of these results, it is recommended to use hybrid models in accident analysis.