TRAFFIC ACCIDENT RISK PREDICTION AND SEVERITY CLASSIFICATION IN THE CITY OF NEW YORK

Abstract

Annually 1.35 million people worldwide die in road traffic, while 20-50 million get injured. That means that every day, almost 3,700 people are killed globally in crashes. More than half of those killed are pedestrians, motorcyclists, or cyclists. In this paper, we analyze the factors that directly affect the risk of motor vehicle crashes and propose the methodology for predicting traffic accidents and classification of collision severity. We were using various machine learning classification algorithms. We present a dataset obtained by extracting accident, road, traffic, and weather-related information from various data sources. The proposed model can identify the time and place when the risk of traffic accidents is higher so that risk can be reduced with proper actions. Experimental results show that the traffic accident prediction model can reach an accuracy of 70%, while the collision severity classification model achieves a macro-average F1-score of 0.56.