Performance of Traffic Accidents’ Prediction Models

Abstract

Modeling traffic-accident frequency is a critical issue to better understand the accident trends and the effectiveness of current traffic policies and practices in different countries. The main objectives of this study are to model traffic road accidents, fatalities and injuries in Jordan, using different modeling techniques, including regression, artificial neural network (ANN) and autoregressive integrated moving average (ARIMA) models and to evaluate the safety impact of travel-restriction strategies during Covid-19 pandemic on trafficaccident statistics for the year 2020. To accomplish these objectives, data of traffic accidents, registered vehicles (REGV), population (POP) and economic gross domestic product (GDP) from 1995 through 2020 were obtained from related sources in Jordan. The analysis revealed that accidents, fatalities and injuries have an increasing trend in Jordan. Root mean of square error (RMSE), mean absolute error (MAE) and coefficient of multiple determination (R2) were sued to evaluate the performance of the developed prediction models. Based on model performance, the ANN models are the best, followed by the ARIMA models and then the regression models. Finally, it was concluded that the strategies undertaken by the government of Jordan to combat Covid-19, including complete and partial banning of travel, resulted in a considerable reduction of accidents, injuries and fatalities by about 35%, 37% and 50%, respectively. KEYWORDS: Traffic accidents, Artificial neural network, Covid-19 pandemic, Regression, Timeseries analysis, Prediction model