To address the issues of low reliability, poor accuracy and insufficient intelligence of the existing coal mine roof rock pressure early warning technology, this study investigated a machine learning-based approach for analyzing roof pressure warning indicators on the working face. Linear regression and systematic clustering methods were employed to analyze the cyclic internal load of the hydraulic support and the periodic weighting step, respectively. A roof pressure analysis and warning model was constructed to enable automatic analysis and prediction of key indicators for roof disaster monitoring on the working face. Field application results demonstrated that the warning system exhibited high reliability, comprehensive analysis and warning functionality, and high accuracy.In the statistical analysis of roof periodic pressure actual occurrences in the demonstration mine, the prediction accuracy was no less than 90%. This system can provide important support for the prevention and control of roof disasters in coal mines.