Hybrid model for detection of severity in car accidents

This research proposes a hybrid deep learning model for predicting car accident severity (minor, moderate, severe) using image categorization. The suggested model extracts features using ResNext50 and improves accuracy with Random Forest. It also evaluates ResNet50 and VGG19 as baseline models. Preprocessing methods such as normalization and augmentation enhance generalization. Adam optimizer was utilized to train the models, and they were assessed using accuracy, precision, recall, and F1-score. The proposed hybrid model performs at 98% accuracy, outperforming existing methods. Confusion matrices are utilized to validate effectiveness of categorization. The method gives automatic severity prediction, leading to faster emergency response. The research puts focus on how deep learning helps to decide on the severity of an accident, and the research indicates that it has promise for use in road safety.