Worldwide, road crashes are a major problem, causing too many deaths, injuries, and hospital stays. They also bring serious harm to public safety and economies. Crashes and injuries are increasing, making, vital to find better methods to predict how severe an accident could be. Good predictions allow for more effective safety actions. Our research used machine learning to forecast crash severity. We looked at various factors like the type of vehicle, weather, road conditions, and traffic levels. We tested several standard models (Random Forest, Decision Tree, SVM, Logistic Regression) to find which performed best for this task to make the model sharper in accuracy. We zeroed in on the most valuable predictive information and stripped away elements that didn't contribute much. Running this new model on real accident data proved it to be 95% accurate. This dependable severity prediction allows us to study patterns of severe accidents. Applications like this provide useful assistance for policymakers and traffic authorities. They can more accurately pinpoint high-risk locations, decide how to use safety resources, and implement practical steps to improve road safety. The study also demonstrates the versatility of machine learning, which could be applied to real-time severity predictions to increase the effectiveness of traffic management. This research shows that machine learning is a practical tool for intelligent transport systems. It provides a data-driven approach to managing traffic, reducing accident risks and their economic cost, and ultimately helps make roads safer for everyone.