INTELLIGENT ACCIDENT SEVERITY CLASSIFICATION USING DATA-DRIVEN MODELS

Abstract

Timely and accurate identification of road accident severity plays a crucial role in reducing human casualties, minimizing property damage, and improving the effectiveness of intelligent traffic management systems. However, the inherent imbalance in accident datasets significantly affects the performance of conventional classification models. To overcome this limitation, this study proposes an Intelligent Accident Severity Classification framework based on data-driven machine learning models. The initial feature set is constructed by analyzing variations in traffic flow parameters associated with accident scenarios. To enhance computational efficiency and reduce redundancy, Factor Analysis (FA) is applied for dimensionality reduction, extracting the most informative latent variables. Subsequently, a Weighted Random Forest (WRF) classifier is developed, where Bootstrap sampling is improved to ensure balanced training data distribution. Additionally, the Matthews Correlation Coefficient (MCC) is utilized to assign adaptive weights to individual decision trees, enabling higher influence for more accurate classifiers during ensemble voting. The proposed FAWRF model is evaluated using standard performance metrics such as accuracy, detection rate, false alarm rate, and AUCROC. Experimental results demonstrate that the proposed approach significantly improves classification performance on imbalanced datasets compared to traditional Random Forest models, achieving robust and reliable accident severity prediction suitable for real-world intelligent transportation systems.