PREDICTION OF RQD AND GSI THROUGH REGRESSION AND MACHINE LEARNING IN ROCK MASSES OF PUNO - PERU

Objective: This study aimed to evaluate the performance of regression-based and machine learning algorithms for predicting the Geological Strength Index (GSI) of rock masses from the Rock Quality Designation (RQD) and discontinuity condition parameters in the Puno region, southern Peru. Methods: Geomechanical data were collected from nine outcrop stations distributed across andesite, volcanic breccia, and arkosic sandstone units belonging to the Puno-Tacaza Group. RQD, joint condition rating (JCond89), and joint spacing were used as predictors in three models: multiple linear regression, support vector regression, and random forest regression, evaluated through the coefficient of determination (R2) and the root mean square error (RMSE) on an independent validation subset. Results: The random forest model achieved the highest predictive accuracy (R2 = 0.94, RMSE = 1.8), outperforming support vector regression (R2 = 0.91) and linear regression (R2 = 0.87). RQD was identified as the most influential predictor, followed by joint condition rating. Conclusion: Machine learning regression, particularly random forest, provides a reliable and reproducible alternative for estimating GSI in heterogeneous andean rock masses, reducing subjectivity associated with conventional chart-based classification.