Machine Learning Based Approach for Predicting the Impact of Time of Day on Traffic Accidents

In India, there is a dearth of information about machine learning (ML) based crash prediction that uses crash data from police crash databases. This study introduces a novel approach utilizing artificial neural network (ANN) to develop models that predict the impact of time of day on road traffic accidents (RTAs) based on Kokrajhar police data. By resourcing a comprehensive crash database spanning from 2020 to 2022, this study encompasses multiple ANN models via different combinations of activation functions and optimizers ensuring enhanced prediction accuracy. The best fit prediction model is found to be produced by the ANN model using the adam optimizer and softmax activation function. Additionally, it was found that the variable "p.m. peak" is the most sensitive factor impacting the crash outcome (fatal/non fatal). Finally, the Apriori algorithm was used to identify the factors that have the greatest impact on each classification of the time of day for fatal traffic accidents. Determining these causal risk variables will facilitate the development of focused mitigation strategies to improve road safety standards.