Miners' unsafe behaviors are the core cause of coal mine accidents, and their risk consequences vary significantly across different work scenarios. A deeper understanding of the relationship between unsafe behaviors and accident types is crucial for achieving refined management of unsafe behaviors. Based on 2,697 coal mine accident reports collected nationwide in recent years, this study identified eight categories of typical unsafe behaviors and nine types of accidents, and constructed three predictive models: logistic regression, random forest, and enhanced random forest, to evaluate accident severity under the interactive effects of unsafe behaviors and accident types. The results show that the enhanced random forest model outperforms the others across five evaluation metrics, including accuracy, precision, recall, F1 score, and AUC, with an overall recognition rate exceeding 85 percent for different accident severity levels, indicating the best comprehensive predictive performance. Among the 72 unsafe behavior–accident type combinations, “violation of safety regulations and water inrush accidents” and “improper command and supervision and gas accidents” were identified as the most representative high-risk combinations, both predicted as major accidents. This finding suggests that coal mine accident risks are shaped by the joint influence of behavioral deviations and situational factors. Furthermore, a risk evaluation mechanism was introduced to quantitatively rank the 72 combinations, providing a theoretical basis for graded warning and differentiated prevention of unsafe behaviors in coal mine enterprises.