Machine Learning Models for Road Accident Prediction for Smart Cities: A Comprehensive Analysis

Abstract

Road accidents are still a prominent urban issue, worsened by a growing population and uncertain weather patterns. Conventional accident prediction models use static historical data, which does not enable them to be responsive to real-time traffic patterns. This research is responding to the demand for predictive models that are adaptive, smart, and can facilitate smart city infrastructure through real-time data integration. Machine learning models—Multilayer Perceptron, Gradient Boosting, Random Forest, Support Vector Machine, and K-Nearest Neighbors—were tested with both historical and real-time traffic data. The models were optimized and trained on varied datasets to improve prediction accuracy. Of these, Gradient Boosting recorded the highest accuracy at 88.1%, followed by Random Forest at 83.73%, showing the power of ensemble learning techniques in predicting accidents. This research emphasizes the significance of real-time data integration for accident prediction and prevention. Merging environmental elements like weather conditions and traffic congestion improves prediction quality, allowing proactive prevention of accidents. By leveraging these models, cities can shift towards data-based road management, infrastructure planning, and congestion control. The results show that real-time models of accident prediction have the potential to enhance urban safety, opening the door to smarter and more efficient traffic management systems.