Authors : Sameh Fuqaha, Ahmad Zaki
DOI : 10.37385/jaets.v7i1.8199
Volume : 7
Issue : 1
Year : 2025
Page No : 679-707
This study develops an interpretable machine learning framework to predict the mechanical properties of bentonite plastic concrete (BPC), an essential material for low-permeability geotechnical structures. Traditional testing of BPC is time- and cost-intensive, while empirical equations often fail to capture the nonlinear effects of bentonite and curing conditions. To address these limitations, four ensemble learning models were optimized using the Forensic-Based Investigation Optimization (FBIO) algorithm, a parameter-free metaheuristic inspired by investigative search processes. The models were trained on three curated experimental datasets to predict slump, tensile strength, and elastic modules. Among all, XGB–FBIO achieved the highest accuracy for slump (R² = 0.98) and tensile strength (R² = 0.99), while GBRT–FBIO performed best for elastic modulus (R² = 0.97). SHapley Additive exPlanations (SHAP) analysis revealed curing time, cement, and water content as the most influential variables. The results demonstrate that the proposed framework can replace repetitive laboratory trials with data-driven insights, providing engineers with a reliable, explainable, and resource-efficient tool for optimizing BPC mix designs in environmental and geotechnical applications.