Machine Learning Applications for Pedestrian Safety in Urban Transportation

Abstract

Pedestrian crossings have a crucial role in urban transportation. The design of pedestrian crossings in accordance with standards enables the reduction of financial losses, air pollution and delays in traffic, as well as improvements in pedestrian safety. In this study, the aim is to examine, using machine learning, whether all pedestrian crossings in a city are designed according to standards. Eleven basic variables were determined for a total of 719 pedestrian crossings in the city centre of Erzurum, Turkey. The problems detected in the pedestrian crossings in the city were divided into nine different classes using these variables. Problems in pedestrian crossings were classified by machine learning algorithms: decision tree, naive Bayes, k nearest neighbours, regression, multilayer perceptron and support vector machine. In classification results, the decision tree algorithm gave more successful results than other algorithms. It was determined that approximately 74% of pedestrian crossings have some problem. According to the results obtained with this algorithm, it was determined that a lack of marking is frequently encountered in pedestrian crossings and that there is an insufficiency of wheelchair ramps. In the last part of the study, suggestions for solutions for all these problems detected in pedestrian crossings are presented.