A Support Vector Machine Approach in Predicting Road Traffic Mortality in Malaysia

Abstract

The traffic mortality rate is the baseline through which road safety plans of a country could be evaluated. A reliable and reasonable analysis of road traffic-related injuries and their leading causes is vital to the road safety investigation, evaluation as well as policymaking. Malaysia has the third highest fatality rate from road traffic accidents in Asia as well as in Southeast Asia. Although many researchers have attempted to provide predictive models of Road Traffic Mortality (RTM) in the country, the predictions are found to be rather unsatisfactory in forecasting the causes as well as the future road fatality. It is hypothesized that the inability of the previous models to provide a good prediction of the RTM might be attributed to the complicated and non-linear data relationship of the underlying causes of road traffic accidents. A Support Vector Machine (SVM) is demonstrated to be effective in solving both classifications as well as regression problems owing to its efficacy to cater for the non-linear relationship of a dataset. The present investigation proposed the application of SVM based model variations namely the Linear, Quadratic, Cubic, Fine, Medium, as well as Coarse Gaussian-based SVM in predicting the RTM. A dataset from 1972 to 1994 was obtained from the Malaysian road traffic database. The data were trained on the SVM model variations. It was demonstrated that the Linear-based SVM model can provide a reasonable prediction of the RTM with only a 12 % error. It is, therefore, inferred that a reasonable prediction of RTM in Malaysia could be achieved through the employment of non-conventional statistical techniques.