Application of support vector machine for crash injury severity prediction: A model comparison approach

Abstract

The study presented in this paper investigated the application of using support vector machine with different kernel functions for crash injury severity prediction. A support vector machine model was developed for predicting the injury severity related to individual crashes based on crash data. The models were developed using the input parameters of driver’s age and gender, the use of a seat belt, the type and safety of a vehicle, weather conditions, road surface, speed ratio, crash time, crash type, collision type and traffic flow. Also, three injury levels were considered as output parameters for this study (i.e. no injury, evident injury and fatality). The overall prediction accuracy of the support vector machine model was compared to the multi-layer perceptron, genetic algorithm, combined genetic algorithm and pattern search. The results demonstrated that the constructed multi-layer perceptron’s performance was slightly better than the support vector machine for injury severity prediction. Whereas, support vector machine involves much lower computational cost than multi-layer perceptron because of using a straight forward algorithm in learning phase. The percent of prediction accuracy for the multi-layer perceptron model was 86.2%, which was higher than the support vector machine model with polynomial kernel (81.4%) followed by the combination of the genetic algorithm and pattern search (78.6%) and genetic algorithm (78.1%). The classification results of the two-level (no-injury and evidence injury/fatality) support vector machine found to be 85.3% was higher than multi-class classification (81.4%). Keywords: Crash Injury Severity Prediction, Genetic Algorithm, Multi-Layer Perceptron, Pattern Search, Support Vector Machine