Forecasting the number of road accidents in Poland and Malaysia using neural networks

Abstract

Road traffic accidents remain a significant concern globally, including in Poland and Malaysia, despite an overall downward trend observed in recent years. While the COVID-19 pandemic has undeniably influenced accident rates, the figures still indicate a pressing need for proactive measures to further reduce their occurrence. This study aims to forecast the number of road traffic accidents in Malaysia and Poland between 2024 and 2030 using neural network models. Historical annual accident data from official national sources were analysed and processed using multilayer perceptron (MLP) neural networks. Two training-validation-testing data splits (70-15-15 and 80-10-10) were employed to assess model performance. The results indicate a likely stabilization in accident numbers in Poland, while Malaysia shows a post-pandemic increase. Despite the limitations of using aggregated data, the models achieved low prediction errors, with MAPE as low as 2.14% in the best configurations. These forecasts can inform evidence-based road safety policies in both countries. The results indicate that the number of road traffic accidents is expected to reach a stable level in the coming years. This outcome appears to be shaped by several key dynamics, including the ongoing development of transport infrastructure—particularly the addition of new highways and express routes—as well as the persistent rise in vehicle ownership. However, it should be emphasized that the reliability of these projections is constrained by the limitations inherent in the random sampling of datasets applied during the model’s training, testing, and validation phases.