Malaysian Road Accident Severity: Variables and Predictive Models

Road accident refers to an incident where at least one land vehicle with one or more people injured or killed. While there are many variables attributed to road accident, ranging from human to environmental factors, the work presented in this paper focused only on identifying predictors that could potentially lead to fatality. In this study, the raw dataset obtained from the Malaysian Institute of Road Safety Research (MIROS) was firstly preprocessed and subsequently transformed into analytical dataset by removing missing values and outliers. Such transformation, however, resort to large feature space. To overcome such challenge, feature selection algorithms were employed before constructing predictive models. Empirical study revealed that there were 26 important predictors for predicting accident fatality and the top five variables are month, speed limit, collision type, vehicle model and vehicle movement. In this work, six predictive models constructed were Random Forest, XGBoost, CART, Neural Net, Naive Bayes and SVM; with Random Forest outperformed the rest with an accuracy of 95.46%.