Predictive modelling on the compressive strength and fresh properties of self-compacting recycled aggregate concrete using machine learning approaches

Self-Compacting Recycled Aggregate Concrete (SCRAC) has emerged as a practical engineering solution to recycle the construction aggregates. This study focusses on developing and comparing six popular Machine Learning (ML) models based on Multi Layer Perceptron regression (MLP), Gradient Boost Regression (GBR), Bagging regression (BG), Extreme Gradient Boost regression (XGB), K Nearest Neighbours regression (KNN), and Support Vector Regression (SVR), along with six metamodels prepared by stacking of these ML models, for prediction of each, Compressive Strength (CS), Slump Flow (SF) and V Funnel time (VF) of the SCRAC. Eight mix ratios were used as the input features. The best performance ranking was exhibited by Gradient Boost model GBRCS for CS prediction whereas stacked models performed best for prediction of SF and VF. Additionally, multi-objective optimization and partial Pareto fronts were constructed to analyse trade-offs between CS, SF, cost, and CO2 emission.