Abstract
This study explores the effectiveness of point load tests (PLI), specifically both diametral (PLI(d)) and axial tests (PLI(a)), in forecasting various rock types' Unconfined Compressive Strength (UCS). Additionally, it examines the implementation of Regression learner app with five machine learning (ML) models to enhance prediction accuracy. These models include Linear Regression, Stepwise Linear Regression, Support Vector Machine, Gaussian Process Regression, and Neural Network. The investigation focuses on rock samples sourced from Pallavaram in Chennai and Panathur in Bangalore. To ensure the reliability of the developed ML models and to assess the best predictive model, performance metrics such as Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Coefficient of Correlation (R²) were employed for validation. The findings suggest that all tested models exhibited commendable performance correlating the parameters through the training and testing phases. Notably, the Neural Network and Gaussian Process Regression models surpassed the performance of the other methods, demonstrating their superior predictive capabilities. The study's results signify a robust correlation between PLI and the UCS of rocks sampled from Pallavaram and Panathur, indicating that the PLI can serve as a valuable metric for anticipating the UCS of rocks. Moreover, an analysis of the correlation across various models revealed that the PLI test performed in the axial direction shows a stronger relationship with UCS, yielding a Coefficient of Correlation (R²) of 0.996, in contrast to the diametral point load index which resulted in an R² of 0.991, applicable for both Pallavaram and Panathur rock samples.