Road Accident Severity Prediction — A Comparative Analysis of Machine Learning Algorithms

Abstract

Crash severity prediction models enable various agencies to predict the severity of a crash to gain insights into the factors that affect or are associated with crash severity. One of the potential ways to predict the crash severity is to leverage machine learning (ML) algorithms. With the help of accident data, ML algorithms find hidden patterns to predict whether the severity of the crash is fatal, serious, or slight. In this research, we develop a prediction framework and implemented six different machine learning algorithms, namely: Naïve Bayes, Logistic Regression, Decision Tree, Random Forest, Bagging, and AdaBoost to predict the severity of the crash. Experimental results procured for the crash dataset published by the UK shows that Random Forest, Decision Tree, and Bagging significantly outperformed other algorithms in terms of all performance metrics. Furthermore, we analyze the huge; traffic data and extract insightful crash patterns to figure out the significant factors that have a clear effect on road accidents and provide beneficial suggestions regarding this issue. We strongly believe that the proposed prediction framework and the extracted pattern analysis would be helpful in improving the traffic safety system and assist the road authorities to establish proactive strategies to prevent traffic accidents.