Effect of Resampling Techniques on Machine Learning Models for Classifying Road Accident Severity in Thailand

Road traffic accidents (RTAs) pose a significant global challenge, particularly in Thailand. This study investigates the impact of resampling techniques on machine learning (ML) models for classifying road accident severity in Thailand, utilizing data from 31,817 road traffic accidents collected between January 1, 2021, and December 31, 2022. The primary challenge addressed is class imbalance, where fatal accidents represent a small fraction of the dataset. Three popular ML models, including Random Forest (RF), K-Nearest Neighbors (KNN), and Extreme Gradient Boosting (XGB), were evaluated with four resampling techniques: Imbalanced (IB), Under-sampling (US), Over-sampling (OS), and Combined Sampling (CS). These resampling approaches generated 12 ML models, whose performance was evaluated under three different train/test split ratios: 70/30, 80/20, and 90/10. Compared to the IB approach, the results demonstrate that all US, OS and CS techniques significantly improved model performance, particularly in terms of F1 score, G-mean, and balanced accuracy. Among the models, RF-CS, KNN-OS, and XGB-CS exhibited the best classification performance. Although these evaluation metrics improved over the imbalanced scheme, KNN’s overall performance in detecting fatal accidents was weaker compared to RF and XGB. Specifically, KNN struggled more with the imbalanced dataset, even after applying resampling techniques. These findings suggest that choosing the appropriate resampling techniques is crucial for enhancing model performance in classifying accident severity.