Compressive strength of bentonite concrete using state-of-the-art optimised XGBoost models

This study proposes an advanced soft-computing approach for predicting the compressive strength (CS) of bentonite concrete using an optimised XGBoost model. Bentonite is valued as a partial cement replacement for its environmental benefits and improved concrete properties, but predicting CS remains challenging due to complex constituent interactions. The study's motivation is the increasing interest in sustainable materials like bentonite as a partial cement replacement, which presents unique challenges due to its high plasticity and swelling properties. While hybrid XGBoost models are effective in civil engineering, their application for CS prediction in concrete is limited. This research simulates hybrid XGBoost models using particle swarm optimisation (PSO), genetic algorithm (GA), and dragonfly optimisation (DO), supported by a comprehensive dataset with varied mix proportions and multicollinearity analysis. Hyperparameter tuning and feature selection techniques were applied to optimise the model's performance. The results demonstrate that the PSO-XGBoost is the best performing model (R2 = 0.974, RMSE = 0.038), followed by DO-XGBoost and GA-XGBoost. All the hybrid XGBoost models perform better than conventional XGBoost model. The developed robust soft-computing based prediction methodology can serve as a reliable alternative tool for predicting the CS of bentonite concrete, thereby facilitating the design and development of sustainable concrete mixtures with enhanced performance characteristics.