RFCNN: Traffic Accident Severity Prediction based on Decision Level Fusion of Machine and Deep Learning Model

Abstract: This research presents RFCNN, a hybrid machine learning and deep learning framework for traffic accident severity prediction using decision-level fusion. The proposed approach combines Random Forest (RF) for feature selection and Convolutional Neural Networks (CNN) for deep feature extraction, followed by ensemble-based classification. The model leverages full and selected feature sets to improve predictive accuracy while addressing challenges like high-dimensional data and class imbalance. Experimental results on real-world accident datasets demonstrate that RFCNN outperforms traditional machine learning models (e.g., AdaBoost, Gradient Boosting, and Voting Classifiers) in terms of accuracy, precision, recall, and F1-score. The system includes a user-friendly GUI for data preprocessing, model training, and performance visualization. The study highlights the effectiveness of feature selection and model fusion in enhancing accident severity prediction, contributing to improved road safety analytics.

Keywords: Traffic Accident Severity Prediction, Machine Learning (ML), Deep Learning (DL), Random Forest (RF), Convolutional Neural Network (CNN), Feature Selection, Ensemble Learning, Decision-Level Fusion, Road Safety Analytics, Predictive Modeling