A hybrid ARIMA-BP approach for superior accuracy in predicting traffic accident losses

Accurately predicting losses resulting from traffic accidents holds crucial significance for accident prevention. Traffic accident forecasting faces challenges. For example, traffic accident forecasting models often exhibit suboptimal accuracy. In this study, these challenges are addressed by integrating the autoregressive integrated moving average (ARIMA) model with the backpropagation (BP) neural network model. Through parameter tuning and model optimization, the predictive accuracy of traffic accident losses is effectively improved, providing more precise forecasting results for accident prevention. Furthermore, a single model may not adequately explore the intricate nonlinear relationships involved due to the complex causal mechanisms of traffic accidents. To address this issue, neural networks are combined with the ARIMA model, and the impact of nonlinear factors in traffic accidents is fully considered. The ARIMA-BP forecasting method is applied to predict the number of traffic accidents, fatalities, injuries, and property damage in China. The results indicate that during the forecast period from 2016 to 2020, the average annual error rates of the ARIMA-BP model for the number of accidents, number of fatalities, number of injuries, and property damage are 4.16%, 3.67%, 7.45%, and 5.94%, with an average prediction error rate of 5.31%. The ARIMA-BP model demonstrated 2.61% and 6.24% lower error rates than the ARIMA model and BP neural network, respectively. The ARIMA-BP model exhibited lower average annual error rates than the individual models. This study substantiates the effectiveness and superiority of the ARIMA-BP prediction model in predicting traffic accident losses, providing a more efficient tool and theoretical basis for assessing and preventing traffic accident risks.