Using AI Model to Anticipate Road Accidents of Poland and Kosovo

This paper investigates the application of artificial neural networks for predicting road traffic accident counts in Poland and Kosovo. The study uses annual accident data collected for the period 2010–2023 from official statistical records provided by the Polish Police and the Kosovo Police. Forecasting procedures were carried out with multilayer perceptron (MLP) models developed in the Statistica environment. To assess the robustness of the models, two alternative dataset division schemes were employed, namely, 70%–15%–15% and 80%–10%–10% for the training, testing, and validation subsets. Forecast accuracy was measured using Mean Absolute Error (MAE) together with Mean Absolute Percentage Error (MAPE). The obtained forecasts suggest that the number of road traffic accidents in both analysed countries is likely to remain relatively stable over the 2024–2030 period. The analysis further demonstrates that a larger share of training observations contributes to improved predictive performance and lower estimation errors. Nevertheless, the relatively short time series constitutes a methodological limitation and requires careful interpretation of the forecasting results.