This study harnesses machine learning to innovate marble waste recycling, delivering a novel, data-driven solution for sustainable construction and industrial applications. Utilising a dataset of 20,000 records, the research pinpointed particle size as a pivotal factor, with finer particles (50 µm) suited for Aggregates. Exploratory data analysis, conducted with precision, revealed significant particle size variation across waste types (ANOVA: F=36.26, p=2.34e-23), guiding meticulous feature engineering, including particle size binning and interaction terms. Three classification models, Random Forest, XGBoost, and Logistic Regression, were rigorously developed, with SMOTE addressing class imbalance. Post-SMOTE, Random Forest achieved a macro-averaged F1-score of 0.52, markedly improving minority class predictions (Calcium Carbonate: 0.49; Other: 0.30), though overall accuracy (0.57) reflects trade-offs in majority class performance. Feature importance and SHAP analyses, clearly presented, underscored Waste Type’s dominance (r=0.683) and particle size’s critical role.
Shambhavi Sinha (Fri,) studied this question.