Machine learning models for estimating contamination across different curbside collection strategies.

Area of Science:

• Environmental Science
• Data Science
• Waste Management

Background:

• Contaminated recyclables hinder circular economy goals and increase processing costs.
• Traditional statistical models have limitations in predicting inbound contamination rates.
• Machine learning (ML) offers advanced methods for contamination prediction.

Purpose of the Study:

• To apply ML models for predicting inbound contamination rates in curbside recycling.
• To compare the performance of different ML models against traditional statistical models.
• To identify key demographic and socioeconomic factors influencing contamination rates.

Main Methods:

• Utilized demographic features from 15 U.S. counties with diverse collection strategies.
• Applied and compared machine learning models including linear regression, support vector machine (SVM), and random forest.
• Evaluated model performance using R-squared and mean absolute error (MAE).

Main Results:

• Machine learning models generally outperformed linear mixed models in predicting contamination rates.
• Support vector machine (SVM) models demonstrated the highest performance (R² = 0.75, MAE = 0.06).
• Population, poverty rate (positive correlation), and median age (negative correlation) were identified as significant predictors.

Conclusions:

• ML models, particularly SVM, offer a robust approach to estimating inbound recycling contamination.
• Understanding key predictors like population and socioeconomic factors is crucial for targeted interventions.
• Improved contamination prediction models are essential for advancing circular economy initiatives.