Study of traffic incident classification using support vector machine
Tan, Jing Mei
Date of Issue2017
School of Civil and Environmental Engineering
Traffic incidents such as accidents, vehicle breakdowns, unattended vehicles, and so on, tends to have an impact on traffic conditions of the roads. It is a non-recurring cause of traffic congestion, which would potentially affect the operational performance and safety issues of the traffic systems. Hence, it is essential to detect and predict traffic incidents so as to alleviate the problem as soon as possible. In this project, Support Vector Machine (SVM) model is explored and used for traffic incident type prediction based on traffic data collected. To do so, real-time traffic data for a period of one week is extracted and retrieved from Land Transport Authority (LTA) of Singapore’s DataMall. Statistical analysis of the one week traffic data is carried out to analyse the percentage of incident counts in accordance to the traffic speed bands, road category and incident types. A selection of SVM kernel functions and parameters will be trained and tested to determine the optimum prediction model in regards to the prediction accuracy. As a result, the outcomes showed that the SVM for radial basis function (RBF) kernel of gamma and C values equal to 100 provides the best prediction accuracy with an average of 90%. Furthermore, the RBF kernel function was found to perform better than the linear kernel function.
Final Year Project (FYP)
Nanyang Technological University