Explainable prediction of healthcare waste generation using hybrid PCA–GPR with SHAP for enhanced environmental protection

The increasing generation of healthcare waste (HCW), together with inadequate management practices, poses risks to public health and environmental safety. This situation highlights the need for robust quantitative approaches to support waste estimation and planning. In this study, machine learning–based regression models, including Support Vector Machine (SVM), Ensemble of Trees (ET), and Gaussian Process Regression (GPR), were applied to estimate HCW generation across Indian states. Principal Component Analysis (PCA) was used to reduce data dimensionality and to construct a hybrid model based on the best-performing individual approach. Model performance was assessed using the root mean square error (RMSE) and the coefficient of determination (R²). Pearson correlation analysis was conducted during the preliminary stage to evaluate relationships among input variables, and statistical significance was verified using a two-tailed test. Sensitivity analysis was performed through nonlinear input selection, resulting in eight model configurations for the model creation. Among the individual models, GPR produced the lowest prediction error, with an RMSE of 0.0048 (GPR-2), outperforming the SVM and ET models. The PCA–GPR hybrid model further improved prediction accuracy, achieving a minimum RMSE of 0.0019. SHapley Additive exPlanations (SHAP) analysis was employed to enhance the effect of the model interpretability, revealing that healthcare capacity indicators and urban-demographic factors are the dominant drivers of HCW generation. The results demonstrate that regression-based and hybrid modelling approaches are effective for estimating HCW generation. The study also indicates that prediction performance can be enhanced by incorporating additional socioeconomic and demographic variables, although data availability and data quality remain key limitations, particularly in regions with incomplete historical records.