Explainable AI and machine learning-based analysis of municipal solid waste generation rate: a South African case...

The upsurge in solid waste generation poses a significant challenge that must be addressed proactively and sustainably. Despite extensive research on the application of machine learning (ML) models in their black-box nature in waste generation forecast, less attention has been given to the multi-factor complexity and multicollinearity, explainability, regional heterogeneity, and latent pattern of waste generation. This study addresses this gap by developing a multi-stage ML-framework integrating principal component analysis (PCA), SHapley Additive exPlanations (SHAP), k-means clustering, and an Adaptive Neuro-Fuzzy Inference System (ANFIS) model. The study utilised a high-dimensional 17 socio-economic, demographic, meteorological, and infrastructural variables in South Africa. The PCA retained about 90.3% of the data variance in 13 principal components (PCs), reducing the complexity of the waste dataset. The k-means cluster unveils 3 distinct waste-generation groups, reflecting different service-access and infrastructure levels. SHAP revealed that access to refuse removal, relative humidity, population density, and household income were the most influential variables to waste generation prediction. The ANFIS model with Grid Partitioning (GP) clustering based on PCA-transformed input outperformed others with RMSE, MAE, MAPE, and R² values of 0.1939, 0.1655, 21.31, and 0.8943 at the training. This research provides intelligent data-driven insights that aid municipal planners in facilitating region-specific and targeted policy formulations, resource and infrastructure allocations and optimization.