Metaheuristic-Enhanced SVR Models for California Bearing Ratio Prediction in Geotechnical Engineering

Abstract

Soil resistance characteristics, particularly the California Bearing Ratio (CBR), play a pivotal role in pavement and subgrade design. However, conventional laboratory-based CBR testing is often time-consuming, labor-intensive, and costly. This study presents a novel machine learning framework that combines Support Vector Regression (SVR) with three recent metaheuristic optimization algorithms—Dingo Optimization Algorithm (DOA), Alibaba and the Forty Thieves Optimization (AFT), and Adaptive Opposition Slime Mold Algorithm (AOSMA)—to predict CBR values efficiently and accurately. A dataset consisting of 220 soil samples with eight geotechnical input parameters was used to develop and evaluate the hybrid models. The predictive performance of each model was assessed using multiple evaluation metrics, including R², RMSE, MSE, RSR, and WAPE. Results indicate that the SVR–AFT (SVAF) hybrid model outperformed the others, achieving an R² of 0.9968 and an RMSE of 0.7946 in the testing phase, demonstrating high generalization ability and predictive precision. The integration of SVR with metaheuristic algorithms significantly enhances model robustness and accuracy, offering a practical and cost-effective alternative to empirical CBR testing methods. This work highlights the potential of hybrid AI models in solving complex geotechnical prediction problems and contributes to the growing body of research at the intersection of civil engineering and artificial intelligence.