Machine Learning Models for Municipal Solid Waste Generation Forecasting: Models, Predictors, Performance, and Future Directions

Abstract

This work addresses the critical research gap of unreliable waste predictions amid rapid urbanization, socioeconomic variability, and external shocks that traditional statistical methods cannot adequately handle by synthesizing studies on machine learning (ML) applications for Municipal Solid Waste (MSW) generation forecasting. Narrowly scoped to quantity (and related composition) forecasting, the review excludes broader waste management aspects such as routing, classification, or lifecycle processes. Its novelty lies in the focused, prediction-centric synthesis that benchmarks models, predictors, and performance metrics while highlighting underrepresented eXplainable Artificial Intelligence (XAI) applications and cross-regional insights. A pre-supplied corpus was filtered using strict inclusion criteria (ML-based empirical forecasting with reported performance metrics) and exclusion criteria (pre-2022 publications or non-forecasting studies), with data extracted on publication trends, geography, waste types, models, predictors, and metrics, then synthesized through narrative review and quantitative summaries. Results show a publication surge peaking in 2025, Asia-dominant research (~60%), MSW focus (~80%), tree-based ensembles (~40%, e.g., XGBoost with R² 0.88–0.95), deep learning (~30%, LSTM R² >0.90 for time-series), and socioeconomic predictors (~60%). While performance benchmarks indicate high accuracy, gaps persist in data scarcity, regional bias, low XAI adoption (~15–20%), and limited transferability. These findings underscore ML’s potential to support SDG-aligned waste optimization, with future directions emphasizing hybrid-IoT frameworks, federated learning, and expanded XAI for more equitable, transparent, and actionable global waste forecasting tools.