Advancing Waste-to-Energy Systems in Institutional Settings: Machine Learning Models for Sustainable Energy Recovery from Municipal Solid Waste at University Campuses

Municipal solid waste (MSW) management in institutional settings, particularly university campuses in developing countries, presents both environmental challenges and energy recovery opportunities. This study integrates empirical characterization data from the University of Jos hostel system with advanced machine learning approaches to develop predictive models for waste generation, net calorific value (NCV), and energy potential. Traditional deterministic methods provide only static estimates, whereas temporal variability and operational uncertainties inherent in academic environments necessitate more sophisticated analytical approaches. Four supervised regression algorithms: linear regression, random forests, gradient boosting, and multi-layer perceptron neural networks were applied to predict daily MSW generation (R2 = 0.9361), NCV, and electrical energy potential. A weighted ensemble model combining all three top-performing algorithms achieved superior cross-validation performance (R2 = 0.8286, RMSE = 133.80 kWh/day) compared to any individual model. Primary data from the University of Jos hostels documented 2.23 tons/day of waste with an NCV of 15.23 MJ/kg, yielding a theoretical monthly electrical potential of 2,776.3 kWh equivalent to approximately 1.8% of institutional electricity consumption. Student population, occupancy rates, and waste composition fractions (food residue and polythene) were identified as the primary predictive features. The proposed machine learning framework enables scenario-based forecasting for facility planning, waste policy evaluation, and optimization of decentralized WTE systems. This methodology is readily transferable to other institutional campuses in sub-Saharan Africa and contributes to advancing sustainable, data-driven waste management practices in resource-constrained environments.