Road traffic accidents pose a significant challenge to socio-economic development, particularly in developing countries where rapid urbanization and increasing traffic demand intensify safety risks. This study aims to evaluate and compare the performance of Weight of Evidence (WoE) and Random Forest (RF) models for traffic accident risk mapping using Geographic Information Systems (GIS) and remote sensing data in Thanh Hoa Province, Vietnam. Multi-source spatial data, including Sentinel-2 imagery, Digital Elevation Model (DEM), road network, and 1183 accident points (2020–2023), were processed to derive key influencing factors such as land use/land cover (LULC), terrain slope, road gradient, intersection density, and accident density. The dataset was divided into training (80%) and validation (20%) subsets, and model performance was evaluated using ROC curves, AUC, Accuracy, and F1-score. The results show that both models effectively capture spatial accident patterns; however, the RF model demonstrates superior performance with AUC = 0.93 and Accuracy = 0.88, compared to AUC = 0.87 and Accuracy = 0.84 for the WoE model. High and very high-risk road segments account for only 14.3% of the total road network but concentrate up to 83.1% of traffic accidents, indicating a strong spatial concentration of accident occurrences. These findings confirm that integrating GIS, remote sensing, and machine learning provides a reliable and effective framework for traffic accident risk assessment, supporting decision-making in traffic safety management and infrastructure planning.