Novel Machine Learning-Driven Road Accident Analysis: A Comparative Study for Predictive Safety and Infrastructure Planning

Abstract

Road traffic accidents remain a critical global safety concern, demanding proactive rather than reactive mitigation strategies. This paper presents a comprehensive analysis of the U.K. Road Accident dataset, leveraging machine learning to predict accident frequency and uncover contributing factors. We perform extensive data preprocessing and feature engineering to transform raw accident records into a structured format suitable for time-series forecasting. A suite of predictive models, including regularized linear models (Lasso, Ridge), Support Vector Regression (SVR), and Facebook’s Prophet, are trained and rigorously evaluated. Our comparative analysis, based on metrics such as Root Mean Squared Error (RMSE), R-squared, and Mean Absolute Error (MAE), demonstrates that models like Lasso, Prophet, and SVR consistently outperform traditional tree-based methods, achieving R² scores of up to 0.99. The findings highlight the efficacy of machine learning in providing robust predictive insights for proactive road safety interventions and data-informed civil engineering practices. This study offers a valuable framework for leveraging historical data to enhance transportation safety and guide future infrastructure development