Data Mining Applied for Accident Prediction Model in Indonesia Toll Road

The current research on toll road accident (TRA) is mainly conducted using conventional descriptive statistics, which, however, fail to properly identify cause-effect relationships and are unable to construct models that could predict accidents. Alternative to decrease traffic accident is by developing accident prediction model. The model relates accident frequencies with traffic flow and various roadway environment characteristics contributing to accident occurrences. This paper presents the TRA prediction model for Jakarta Outer Ring Road Toll Road (JORR), to identify the most important causes of accidents and to develop predictive models. Data mining (DM) techniques (artificial neural networks (ANNs) and support vector machines (SVM)) were used to model accident and incident data compiled from the historical data. Based on the R-Tools, results were compared with those from some classical statistical techniques (logistic regression (LR), revealing the superiority of ANNs and SVM in predicting and identifying the factors underlying accidents in toll road.