Evolutionary symbolic models for predicting mortar strength incorporating plastic waste and supplementary cementitious materials for sustainable construction

This study used cutting-edge machine learning methods to create models that could predict the compressive strength (C-S) of plastic-based mortar (PBMs). The methods used included Gene Expression Programming (GEP) and Multi-Expression Programming (MEP). Compared to more conventional black-box machine learning techniques, evolutionary symbolic regression models have the benefit of producing clear mathematical equations that reveal the connections between input parameters and output strength. The models were trained and validated using a large dataset that included various PBMs mix proportions. Hyperparameter tuning was performed to optimize model performance and improve predictive accuracy. The MEP model achieved a higher R² value of 0.940 than the GEP model, which was at 0.913. The predictive capabilities of both models were further validated using statistical performance metrics and cross-validation using Taylor's diagram. A rigorous sensitivity analysis was conducted to quantify the influence of key mix components on C-S. The findings of this study provide valuable insights into optimizing PBMs formulations for enhanced C-S, reducing dependency on conventional cement, and promoting sustainable construction practices. The results also establish the effectiveness of GEP and MEP as reliable predictive tools for evaluating and improving PBMs performance in real-world applications.