Traffic Accident Prediction Using Machine Learning Based on PT Jasa Raharja Data

Abstract

Traffic accidents represent a critical issue that significantly affects public safety and generates substantial social and economic impacts, particularly within the operational area of PT. Jasa Raharja Lhokseumawe Branch. The lack of predictive information regarding accident occurrences often results in reactive policy making. This study aims to develop a machine learning–based forecasting model for traffic accident rates using a combination of K-Means Clustering and Recurrent Neural Network (RNN). The dataset consists of historical traffic accident records from 2022 to 2024, which were preprocessed and aggregated on a weekly basis at the district level. K-Means Clustering was employed to group districts according to weekly accident patterns, resulting in two optimal clusters based on silhouette score evaluation. Subsequently, separate RNN models were developed for each cluster to forecast weekly accident occurrences. Model performance was evaluated using Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE). The results indicate that the RNN model achieved higher prediction accuracy for clusters with more stable accident patterns compared to clusters exhibiting higher fluctuation. Overall, the proposed combination of clustering and RNN demonstrates strong potential in producing accurate traffic accident forecasts