Accidents that happen on the road in Bangladesh are very challenging to the life and financial integrity of the people. The proposed study is a machine learning framework that can be used to predict the severity of accidents using past and recent accident data. We used the information from the Accident Research Institute (2005-2015) and the Bangladesh Road Transport Authority (2023-2024). The models trained were Decision Tree, Random Forest, Balanced Random Forest, XGBoost, and Naive Bayes, with SHAP values used to interpret them. To be robust, the cross-validation was used, and the assessment was guided by category-specific confusion matrices. The Decision Tree model scored higher in the mean of F1-score than others, with a score of 0.871. To be usefully accessible, a Streamlit web application that reproduces real-time predictions was created. This study will help a policymaker understand the level of accident risks during times of day, types of vehicles, and weather conditions, to enable actionable intelligence.