This study presents a comprehensive, harmonized dataset and an interpretable analytical framework for assessing per-capita waste generation (PCWG) across 36 Indian states and union territories. The research employs a robust data preprocessing methodology combined with a hybrid feature-selection approach, incorporating mutual information, tree-based importance metrics, and SHAP-based explainability, ensuring a transparent identification of key influencing factors. Supervised machine-learning models, including CatBoost and Random Forest, were trained to capture complex nonlinear relationships in municipal solid waste data, achieving high coefficients of determination (R2) of 0.952 and 0.802, respectively. Explainability analyses consistently highlight urbanization intensity and population density as critical drivers of PCWG variability. For scenario exploration, the predictive models are integrated with Bayesian optimization and NSGA-II-based multi-objective optimization to assess trade-offs between waste generation, treatment capacity, and implementation costs. The resulting Pareto-optimal solutions, along with annotated datasets, offer a valuable resource for comparative analysis, policy development, and urban sustainability studies. All data, workflows, and model outputs are designed to ensure reproducibility and transferability in diverse regional and environmental management contexts.
Neelemagam et al. (Thu,) studied this question.