Road safety remains a critical global issue, particularly in densely populated urban environments where traffic dynamics are highly complex. Predicting and preventing accidents is challenging due to the interplay of multiple contributing factors and the need to integrate diverse data sources such as spatial, temporal, and real-time traffic information. This study presents an integrated framework for jointly predicting road accident rates and fatalities by integrating 37 spatial, temporal, behavioral, and accident-specific dimensions through a combination of machine learning and statistical models. To identify the key determinants of accident frequency and severity, we incorporated geographic mapping, weather, and contextual variables into clustering techniques and a Decision Tree Regressor. This approach enabled the identification of high-risk areas, critical risk factors, and patterns underlying accident occurrence, providing data-driven insights for targeted interventions. An empirical analysis using a real Bangkok dataset of 10,366 accident records demonstrates that combining machine learning methods with spatial and behavioral features derived from statistical models significantly improves predictive accuracy. The proposed framework shows broad applicability across national contexts and offers valuable evidence to design targeted accident reduction strategies. By integrating predictive analytics with geographic mapping, this study underscores the importance of data-driven approaches in advancing road safety and demonstrates clear application value for policy formulation, aligning with the priorities of the World Health Organization for traffic safety.