Ensuring the stability of underground excavations is a major concern in mining engineering, given its direct impact on the safety and efficiency of operations. The dynamic nature of the subsurface environment creates a continuous necessity to explore innovative methods to improve the overall stability of open stopes and eliminate potential risks. This research focuses on advancing the accuracy of underground excavations stability predictions in mining engineering by optimizing an artificial neural network model. Analyzing Potvin's database, which consists of 175 historical cases, we explored the impact of different ANN model configurations; it was discovered that normalizing the data with Standard Scaler and implementing Swish as the activation function in all layers produced the most accurate predictions for this specific case. Furthermore, employing the SHAP (Shapley Additive exPlanations) tool allowed us to analyze the importance of the features and determine the factors with the highest influence. Our findings reveal that the shape factor has the most significant impact on the stability of the underground openings, followed closely by the Q value. This research contributes to the optimization of predictive models for underground mining excavations stability and reveals the critical role played by specific parameters impacting the stability of open stopes.