Severity prediction of motorcycle crashes with machine learning methods

Abstract

This study uses classification algorithms to establish models to predict the severity of crash injuries when motorcycle crashes occur. In this study, the power of multi-layer perceptron (MLP), rule induction (PART) and classification and regression trees (SimpleCart) models for predicting the severity of motorcycle crash was evaluated by comparing their results. To accomplish this objective, motorcycle crash data set extracted from the National Road Traffic Crash Database at the Building and Road Research Institute in Ghana. The data set was classified into four injury severity categories: fatal, hospitalised, injured and damage. The data collected from this database will provide means to directly compare and rank the data mining models, while also allowing for the identification of variables that are significantly influencing the severity of motorcycle crash. The results showed that among the tested classification algorithms, the SimpleCart model with an average accuracy of 73.81% outperformed the PART model (73.45%) and the MLP (72.16%) model based on a 10-fold cross-validation approach. The results revealed that the most significant factors associated with motorcycle crash injury severity were location type, settlement type, time of the crash, collision type and collision partner.

Keywords:

• Motorcycle
• motorcycle crash
• injury severity
• data mining
• classification techniques
• road safety

Acknowledgments

The authors thank the staff of the Traffic and Transportation Division of the Building and Road Research of Council for Scientific and Industrial Research (BRRI-CSIR), Ghana, for providing the data for the study.

Disclosure statement

No potential conflict of interest was reported by the authors.