Multi-Layer Perceptron Approach for Predicting Road Traffic Accidents Based on the Driver Age

Abstract

Road accidents are grown to be a huge disaster over the globe despite geographical boundaries. It is challenging to eradicate the road accident occurring rate but necessary countermeasures can be taken to reduce the accident rate. The driver of the vehicle can be considered as the main responsible person for the accident. In Sri Lanka most of the times drivers try to escape from the accident location. To take necessary legal actions, identifying the driver of the accident is the main consideration. Therefore, the objective of this study is to identify the age group of the driver by using several road accident factors, which can be successfully applied for driver identification. The researcher has used 3 attributes; Direction of vehicle moving, a human crash factor related to the accident and the location of the accident (nearest km post) to identify the driver age group. This research develops a Multilayer Perceptron (MLP) approach to identifying the driver age group with considerable accuracy. Decision tree approach has used to compare the results of the MLP model and the MLP outperformed with 85.52% accuracy. Sensitivity analysis by using essential hyper-parameters was performed in order to identify the best MLP model. For the prediction, analysis results revealed that the 140 epochs with RMSprop optimizer can increase the accuracy of the model and the low learning rates optimized the accuracy of the model. The developed model will be effectively used for identifying driver age limit and the model will be able to use by relevant authorities to make accurate decision making towards the accident investigations.