Forecasting the Intensity of Road Traffic Collisions using Machine Learning Algorithms

• Published in: 2025 8th International Conference on Computing Methodologies and Communication (ICCMC) pp. 795 - 802
• Main Authors: Rithickroshan, M., Deepika, T., Madhana, R., Kesavan, R., Saravanakumar, R., Lakshmi, S. Jeya
• Format: Conference Proceeding
• Language: English
• Published: IEEE 23.07.2025
• Subjects: Accidents; Accuracy; Ensemble learning; machine learning; Machine learning algorithms; Multilayer Perception; Naive Bayes; Predictive models; Real-time systems; Road safety; Road traffic; road traffic collisions; Stacking; stacking ensemble; Supervised learning; supervised learning algorithms; XGBoost
• DOI: 10.1109/ICCMC65190.2025.11140733

Unevenly burdening poorer countries, road traffic collisions cause a great deal of harm, including injuries, deaths, and economic losses. This situation has been the subject of prior research at various locations and intersections using a variety of approaches. Using data collected from a driving simulator, this study introduces an ensemble machine-learning method for assessing and predicting traffic events. XGBoost (XGB), Naive Bayes (NB), and Multi-layer Perceptron (MLP) are supervised learning algorithms that are included in the methodology. The software uses the stacking ensemble method to predict traffic accidents by combining the results of the base layers. Ensemble methods, such as XGB, NB, and MLP often outperform single-model machine learning approaches in terms of accuracy. These findings have important implications for risk management and traffic safety because they shed light on the factors that contribute to accidents and how to measure their severity. Based on the user's input, the model will determine the likely impact of an accident at their chosen site. Future traffic accident reduction efforts will owe a great deal to this model, which will play an essential role in roadway design and administration. To improve performance in forecasting road traffic collision intensity using XGB, NB, and MLP with ensemble learning, optimize hyperparameters, ensure quality feature selection, balance class distribution, use cross-validation, and apply stacking or voting ensembles to leverage model strengths. Incorporate real-time data and spatial-temporal features for enhanced accuracy.