Comparative analysis of machine learning and traditional methods for rock strength prediction

Uniaxial compressive strength (UCS) is a fundamental rock geomechanical parameter widely utilised in various rock engineering applications, including mining, tunnelling, dam construction, and rock slope stability assessments. However, obtaining high-quality core samples for UCS testing is often impractical and costly, requiring indirect estimation methods using parameters such as porosity (n), Schmidt hammer number (SHN), P-wave velocity (V_p), and point load index (Is ₍₅₀₎). Existing empirical relationships, while useful, are typically limited to specific rock types and may lack accuracy across diverse conditions. This study investigates the applicability of machine learning algorithms, support vector regression (SVR), random forest (RF), and artificial neural network (ANN), to predict the UCS of various rock types, including Claystone, Granite, Schist, Sandstone, Limestone, Slate, and Dolomite, using a dataset of 162 samples from the literature. To evaluate the performance of these models, a 10-fold cross-validation method was employed. The correlation coefficients (R²) of the SVR, DT, and ANN models were found to be 0.98, 0.97, and 0.99, respectively. A comparative analysis was conducted considering six traditional empirical models, which yielded R. values in the range of 0.31 to 0.51. The results demonstrate that machine learning models significantly outperform traditional empirical models, indicating their superior capability for UCS prediction. These findings highlight the potential of machine learning techniques to enhance the accuracy and efficiency of rock engineering applications, provided a high-quality dataset is available.