Research on predicting the risk level of coal mine roof accident based on machine learning

Coal mine roof accidents are one of the main types of accidents leading to the decline of coal mine safety productivity, accounting for about 20% of the total number of coal mine safety accidents on average each year. In order to safeguard the safety and health of the workers and reduce the probability of the decline of productivity, the coal mine risks should be made known and controlled by certain technical means. This study proposes and constructs a novel and practical method for predicting the risk level of coal mine roof accidents. The method aims to provide risk early warning, refined management, and dynamic strategy adjustments for accident prevention, thereby improving control effectiveness and ensuring safe coal mine production. Firstly, 379 coal mine roof accidents data were initially collected, and after screening and filtering, an analysis dataset containing 305 cases was established, and a list of attributes describing coal mine roof accidents was constructed; secondly, the high-dimensional and complex roof accident data were downscaled using principal component analysis. Finally, KNN, SVM and DT algorithms are used to evaluate the model performance. Through comparison of model performance evaluation metrics, the Random Forest integration algorithm is introduced to improve the evaluation and prediction of the model, and the prediction accuracy jumps to 0.94. Compared with the previous methods, the recall rate and F1 score also achieve a significant improvement to 0.87 and 0.89, respectively, which is significantly superior to other model performances. The results show that the method proposed in this paper can also be applied to the risk level prediction of other coal mine accidents, assisting coal mine operators in checking safety issues and taking precautions through large-scale data on related safety accidents.