Predicting workability and mechanical properties of bentonite plastic concrete using hybrid ensemble learning

Area of Science:

• Civil Engineering
• Materials Science
• Environmental Engineering

Background:

• Heavy metal contamination in wastewater necessitates efficient and cost-effective remediation solutions.
• Bentonite Plastic Concrete (BPC) shows potential for heavy metal adsorption but faces experimental validation challenges due to cost and complexity.

Purpose of the Study:

• To develop accurate predictive models for BPC workability and mechanical properties (slump, tensile strength, elastic modulus).
• To overcome the limitations of traditional experimental methods in BPC research.
• To provide a user-friendly computational tool for real-time application in civil engineering materials research.

Main Methods:

• Hybrid ensemble learning models (Random Forest, Adaptive Boosting, Extreme Gradient Boosting, Gradient Boosting Regression Tree) were employed.
• Models were optimized using Forensic-Based Investigation Optimization (FBIO).
• Shapley Additive Explanation (SHAP) analysis was used to identify key influencing parameters.

Main Results:

• The Gradient Boosting Regression Tree optimized with FBIO (GBRT-FBIO) achieved the highest accuracy in predicting elastic modulus (E).
• The Extreme Gradient Boosting optimized with FBIO (XGB-FBIO) model demonstrated superior performance for predicting tensile strength (TS) and slump (S).
• SHAP analysis revealed water as the most significant factor for slump, and curing time as the most critical factor for tensile strength and elastic modulus.

Conclusions:

• Hybrid ensemble learning models optimized with FBIO provide a reliable and cost-effective computational approach for predicting BPC properties.
• These models significantly reduce the need for extensive and expensive experimental testing.
• The developed online tool facilitates the practical application of these predictive models in civil engineering, advancing materials research.