Forecasting heterogeneous municipal solid waste generation via Bayesian-optimised neural network with ensemble learning for improved generalisation

Future projections of municipal solid waste (MSW) generation trends can resolve data inadequacy in formulating a sustainable MSW management framework. Artificial neural network (ANN) has been recently adopted to forecast MSW generation, but the reliability and validity of the stochastic forecast are not thoroughly studied. This research develops Bayesian-optimised ANN models coupling ensemble uncertainty analysis to forecast country-scale MSW physical composition trends. Pearson correlation analysis shows that each MSW physical composition exhibits collinearity with different indicators; therefore, the MSW should be forecasted based on its heterogeneity. The Bayesian-optimised ANN models forecast with smaller relative standard deviations (3.64-27.7%) than the default ANN models (11.1-44,400%). Malaysia is expected to generate 42,873 t/d of MSW in 2030, comprising 44% of food waste. This study provides a well-generalised ANN framework and valuable insights for the waste authorities in developing a circular economy via proper waste management.

Automated summary

This research develops a Bayesian-optimised artificial neural network (ANN) model coupled with ensemble uncertainty analysis for forecasting country-scale trends in municipal solid waste (MSW) physical composition. The Bayesian-optimised ANN models provide more reliable and accurate predictions with smaller errors compared to default models.