PREDICTION OF ROOF FALL IN UNDERGROUND MINE

Roof falls in underground mines represent an extensive risk to worker safety and operational efficiency. These occurrencesfrequently result from rock mass instability, inadequate support, and complex geological conditions. Accurate prediction of rooffalls is critical to mitigating risks, preventing accidents, and ensuring continuous, safe mining operations. Traditional methodsfor predicting roof falls rely heavily on empirical observations and geotechnical data analysis, often lacking precision in dynamicmining environments. In this study, we focus on utilizing deep learning algorithms to predict roof fall incidents more accurately.Deep learning models excel at capturing complex, non-linear relationships among various mining parameters, making them wellsuited for this task. We implemented several deep learning architectures, including MultiLayer Perceptron (MLP), Long Short-Term Memory (LSTM), and Recurrent Neural Network (RNN), to analyze geotechnicaland environmental dataFurthermore, hybrid models enhance the predictive power of standalone algorithms. Specifically, we combined LSTM withattention mechanisms to focus on critical features in sequential data and integrated LSTM with RNN to capture both short-termand long-term dependencies. These deep learning-based approaches significantly improve the accuracy and reliability of roof fallpredictions, providing valuable insights for proactive safety management in underground mining environments.