Machine learning-based prediction of global solid waste generation and composition.

Area of Science:

• Environmental Science
• Data Science
• Urban Planning

Background:

• Rapid urbanization and economic growth exacerbate municipal solid waste (MSW) generation globally.
• Accurate MSW generation and composition prediction is vital for effective waste management and resource recovery.
• Data heterogeneity and complexity hinder reliable global MSW forecasting.

Purpose of the Study:

• To develop and evaluate a data-driven framework for forecasting MSW generation and composition across 217 countries.
• To compare the predictive performance of artificial neural networks (ANN) and multi-linear regression (MLR) models.
• To identify key socioeconomic and demographic drivers of MSW generation.

Main Methods:

• Utilized multi-linear regression (MLR) and artificial neural networks (ANN) for MSW forecasting.
• Incorporated socioeconomic parameters (GDP, population, literacy rate, urbanization, household size) into the models.
• Validated model performance using R² values for prediction accuracy.

Main Results:

• Artificial neural networks (ANN) demonstrated superior predictive accuracy (R²=0.94) for total MSW generation compared to MLR (R²=0.57).
• Existing global models reported lower accuracy (R²=0.68) for MSW generation prediction.
• Predicting waste composition remained challenging (R² up to 0.15) due to unaccounted factors.

Conclusions:

• The proposed data-driven framework, particularly ANN, offers a robust approach for forecasting MSW generation.
• Challenges in predicting waste composition highlight the need for further research into behavioral and regional influences.
• Findings support policymakers in developing sustainable and circular waste management systems aligned with Sustainable Development Goals.