Comparison and Validation of Regression Techniques for Dilution Prediction in Underground Mining

The stability of stopes in underground mining is a critical factor for both operational safety and the economic viability of mining projects. Empirical methods, such as Potvin's Stability Graph (1988), are widely used to predict stability conditions; however, their applicability may be limited in distinct geomechanical scenarios. In this context, this study proposes a quantitative approach based on multivariate regression for predicting dilution throughout the development of different stopes in underground mining. To achieve this, various regression techniques were tested, including linear, polynomial, exponential, and logarithmic models, aiming to predict unplanned dilution based on 18 independent variables related to operational, geotechnical, and structural parameters. The results indicated that second-degree polynomial regression exhibited the best performance, achieving a coefficient of determination (R2) of 82.6% and an absolute mean error below 4%. Statistical analysis revealed that variables such as vertical stress, stope depth, and specific drilling index significantly influence dilution prediction, highlighting the need for an integrated approach between geomechanical and operational factors. Thus, the findings of this study underscore the importance of developing methodologies tailored to local geomechanical conditions, contributing to the optimization of underground mining planning, the reduction of operational losses, and the improvement of extraction efficiency.