Traffic accidents affect public safety, traffic, and economic efficiency, posing serious problems for urban areas. To improve traffic safety and accident prediction, this study investigates the use of Machine Learning (ML) techniques within the context of smart cities. The Study examines accident severity and pinpoints high-risk areas using Random Forest and Logistic Regression models on the US Accidents dataset (2017–2023). While Random Forest captures intricate interconnections for substantial prediction accuracy, Logistic Regression provides interpretability by emphasizing the influence of individual elements. The algorithms use contextual and environmental elements to enhance accident prediction, including weather, road visibility, and regional characteristics. The results demonstrate that AI-powered smart city solutions can reduce traffic risks by enabling proactive measures. Specifically, the Random Forest model achieved an accuracy of 94.1% in predicting accident severity, while Logistic Regression provided interpretable insights into contributing factors (e.g., weather and visibility). These findings allow urban planners to prioritize high-risk areas, optimize traffic management, and deploy emergency resources more efficiently, ultimately promoting safer urban transportation.