Utilization Of Machine-Learning-Based Model Hybridized With Meta-Heuristic Frameworks For Estimation Of Unconfined Compressive Strength

Unconfined compressive strength (UCS) is one of the rocks' most valuable mechanical properties in constructing an accurate geo-mechanical model. It has traditionally been determined through laboratory core sample testing or by analysis of well-log data. After a great deal of effort and growing investment in time, the proper adoption of machine learning methods, especially the radial basis function (RBF), opens a route to promising alternatives against empirical methods for better real-time prediction of UCS. The current study considers the RBF-based machine learning model, whose parameters have been optimized using two enhanced meta-heuristic frameworks: Improved Arithmetic Optimization Algorithm (IAOA) and Flying Foxes Optimization (FFO). Based on an extensive dataset already used in previous studies and applying some soft computing techniques, vigorous performance metrics such as RMSE, R-2, MAE, U95, and MNB were used to test the developed frameworks. The outcomes indicate a significant outperformance of the hybrid RBFF technique over the solo RBF and RBF-IA frameworks. Specifically, the RBFF model resulted in an R-2 of 0.998, an RMSE of 1.313, and an MNB of -0.003, reflecting its better performance in UCS prediction. This study indicates the efficiency of integrating RBF with meta-heuristic optimization to enhance UCS predictions in geotechnical studies.