Comparative analysis of AI models for solid waste forecasting

The accurate prediction of solid waste (SW) generation is a critical component of sustainable waste management, particularly in the face of increasing environmental and socio-economic pressures. Traditional approaches such as surveys, time series analyses, and regression techniques often struggle to capture nonlinear patterns and adapt to changing conditions. This study provides a comprehensive review and comparative analysis of artificial intelligence (AI)-based predictive models applied to diverse waste generation streams. Models examined include hybrid and ensemble frameworks, as well as individual approaches such as SVMs, NARX, Extreme Gradient Boosting (XGBoost), grey relational analysis-Bidirectional Long Short-Term Memory (GRA-BiLSTM), multivariate adaptive regression splines (MARS), and Inclusive Multiple Model (IMM). Unlike previous reviews, we explicitly bridge the gap between algorithmic precision and real-world decision-making, while addressing the unique data-reproducibility challenges found in developing-country contexts. Hybrid and ensemble models consistently achieve superior accuracy, robustness, and generalizability. XGBoost and chaotic frameworks also demonstrate improved computational efficiency and predictive stability compared to traditional black-box models. Importantly, no single model is universally optimal; selection should consider dataset quality, forecasting horizon, and specific application objectives. Hybrid architectures integrating preprocessing, optimisation, and ensemble learning offer the most reliable tools for high-performing and interpretable forecasts. These insights guide researchers, practitioners, and policymakers in developing advanced waste forecasting systems to support informed decision-making and sustainable waste management planning.