Road traffic crashes are a significant public health concern and a major contributor to illness and mortality, with low- and middle-income countries such as Ethiopia bearing a disproportionately high burden. Evaluating temporal trends and predicting road traffic crashes are crucial for implementing effective proactive prevention measures. This study aimed to analyze temporal trends and forecast road traffic crashes in the Amhara region of Ethiopia using a time-series approach. The study utilized secondary traffic crash data from the Amhara region collected by the Amhara Region Police Commission Traffic Department spanning from 2016 to 2024. A univariate Autoregressive Integrated Moving Average (ARIMA) framework was applied to capture underlying trends and seasonality, while machine-learning models, including Random Forest (RF) and Artificial Neural Networks (ANN), were employed to address nonlinear dynamics. Among the classical models, ARIMA (2,1,3) model was identified as the best fit for predicting total traffic crashes, and diagnostic tests, including the autocorrelation function and the partial autocorrelation function, were conducted. The model exhibited robust forecasting performance, achieving a Mean Absolute Percentage Error (MAPE) of approximately 15%, which corresponds to a forecast accuracy of 85%. The results indicate a rising trend in road traffic crashes until 2018, followed by a significant decline thereafter, likely influenced by mobility restrictions during the COVID-19 pandemic and enhanced road safety enforcement. Machine learning models, particularly Random Forest, demonstrated superior performance over ARIMA and hybrid ARIMA-RF models in short-term forecasting, effectively capturing nonlinear patterns and abrupt variations. The findings underscore the complementary roles of time-series and machine learning models in traffic safety analysis, offering valuable insights for targeted interventions and proactive traffic management in the Amhara region.
Fetene et al. (Thu,) studied this question.