The accurate measurement of Uniaxial Compressive Strength (UCS) and Tensile Strength (TS) is crucial in geotechnical engineering, mining, and construction, where it supports data-driven decision-making, enhances structural safety, and optimizes resource utilization. This study advances UCS and TS measurement by integrating advanced deep learning models, including Convolutional Neural Networks (CNN), Gated Recurrent Units (GRU), Long Short-Term Memory (LSTM), and Recurrent Neural Networks (RNN). These models effectively capture complex nonlinear relationships within the input data, ensuring reliable and robust geomechanical property measurement and prediction. A comprehensive measurement-focused evaluation framework was employed, incorporating statistical metrics, radar plots, Taylor diagrams, and the Anderson-Darling (AD) test to critically assess model performance. Partial least squares regression and probability plots were used for data analysis, while SHapley Additive exPlanations (SHAP) analysis and importance value functions provided insights into the influence of key parameters, including P-wave velocity, density, cohesion, and angle of internal friction. The findings identify the LSTM model as the most effective for both UCS and TS measurement and prediction, with CNN as a strong alternative. The LSTM model achieved impressive prediction results for UCS and TS, with coefficient of determination (R2) values of 0.998 and 0.997, respectively. These high R2 values underscore the outstanding performance of the LSTM model in accurately predicting rock strength properties in this study.
