Design and Implement Machine Learning Based Predictive Analytics for Road Safety and Accident Prevention

Coded traffic accident analysis is critical for improving road safety, optimizing emergency response, and enabling smart transport systems due to the rapid rate of urbanization and the persistent increase in traffic quantities. This study delves into the use of ML models to forecast the severity of accidents using a massive dataset of US occurrences from 2016 to 2023. An ordered preprocessing pipeline was adopted, with missing value management, outlier elimination, categorical encoding, designed temporal and environmental variables and Synthetic Minority Over-Sampling Technique (SMOTE) to resolve the imbalance of classes. The AdaBoost and random Forest (RF) models were trained and optimized using 5-fold cross-validation to make them more generalizable and resilient. RF model was best with the greatest accuracy of 97.82% and the greatest precision of 99%, but AdaBoost also gives good performance of 95.67% accuracy and 96.56% recall. Both models performed better than the simple models like XGBoost, CatBoost, and Support Vector Machine (SVM). The analysis of feature significance revealed that variables linked to location, time, and weather were the most relevant predictions for policymakers to understand. The findings prove the relevance of ensemble learning to proactive road safety management, infrastructure planning and accident prevention. This paper connects machine learning with real-world traffic dynamics, which can be used in the future to integrate into intelligent transportation systems.