Analysis and Predictive Modeling of Traffic Incidents in Karachi using Machine Learning

Road traffic accidents have accounted to extremely dense road traffic and the relatively great freedom of movement given to drivers. Due to the increasing traffic accidents in Karachi, it is vital to investigate the major parameters that are causing these fatalities. For this purpose, machine learning techniques provide a greater advantage over other statistical methods. In this research, a novel approach that applies Random Forest and Support vector machine (SVM) algorithm out of many different machine learning algorithms for modeling traffic accidents prediction. Empirical results show that reasonable accuracy of the developed model. The results further showed the accuracy fluctuated according to the number of attributes in the output parameter. The results of SVM showed better predictions than that from Random Forest. The parameter with less attributes like Disposal has higher accuracy of prediction with Random Forest 83.12% whereas those with greater number of attribute have higher prediction accuracy with SVM e.g. Months with 64.98%.