Accident Severity Forecasting Using SMOTE and Ensemble Learning

Abstract:- Traffic accidents remain a critical issue worldwide, posing substantial risks to public safety and infrastructure. This study presents a predictive model aimed at forecasting accident severity, utilizing a dataset comprising temporal, demographic, and vehicle-related features. In order to preprocess the data, missing values are addressed, categorical variables are encoded, and SMOTE is used to handle class imbalance. The model employs an ensemble learning approach, combining RandomForest and GradientBoosting classifiers through soft voting.Hyperparameter optimization is performed using GridSearchCV. The resulting model achieves a high accuracy of 96.41%, with particularly strong performance for severe accidents. However, the recall for minor accidents needs further refinement. This research demonstrates the potential of ML techniques in enhancing road safety through effective accident seriousness prediction.