Analysing Risk Patterns of Road Accidents Severity using ML

Despite being avoidable, traffic accidents continue to be a leading cause of deaths and severe injuries worldwide. In order to predict fatality severity categories (Fatal, Serious, or Slight) for events and recognise accident-prone highway segments based on past data trends, this study presents a machine learning framework. The system gets three kinds of information: accidents, cars, and deaths. Then it cleans up the data, adds features, encodes categories, and log-transforms the data before running it through many models. We use six ML methods and compare them: Logistic Regression (LR), Logistic Regression CV(LRC), Decision Tree (DT), Tuned Decision Tree (using Grid Search), Random Forest, and a Random Forest that has been optimised using GridSearchCV. Exploratory Data Analysis reveals that accidents are common across locations and times, and among different classes of people. Among the different models you have in your hand, Tuned Decision Tree and LRC give the best prediction accuracy, 95%. Feature importance analysis indicates that Engine Capacity, Age of Driver, Age of Vehicle, Day of Week, and Light Conditions are the most influential factors in determining how bad something is. For such severity predictions from various geographic locations, safety measures can be prioritized by checking the road zones where accidents are more likely to occur.