A Comparative Study of Machine Learning Models for Accurate E-Waste Prediction

• Six Machine Learning (ML) models are compared and evaluated for e-waste prediction. • Seven types of e-waste are considered with the multiple ML model prediction models. • A case study of the e-waste inventory project in Malaysia is presented. • ML models are evaluated via Root Mean Squared Error, Mean Absolute Error, and R 2 . • The analysis demonstrates the power of using the GPR model for e-waste prediction. The rapid growth of electrical and electronic equipment waste (e-waste) presents a major environmental challenge. Traditional linear production models fail to optimize resource recovery, while circular economy (CE) strategies remain underutilized due to inadequate forecasting methods. Given the high-value materials in e-waste, accurate prediction is crucial for efficient recycling and regulatory planning. This study evaluates six Machine Learning (ML) models, Linear Regression, Regression Tree, Support Vector Regression, Ensemble Regression, Gaussian Process Regression (GPR), and Artificial Neural Networks, for e-waste forecasting. Using historical Malaysian e-waste data, models were trained and optimized in MATLAB, with performance assessed via RMSE, MAE, and R². Results show that GPR significantly outperforms all models, providing the highest accuracy and lowest prediction errors across all e-waste categories. Optimal GPR kernel functions vary by waste type, with Rational Quadratic, Squared Exponential, Matérn 5/2, and Exponential kernels producing the best results. SVR and ensemble trees also perform well but are slightly less reliable. These findings highlight the need for ML-driven forecasting in e-waste management policies. Integrating GPR into national recycling programs and Extended Producer Responsibility (EPR) policies can optimize collection systems and resource recovery. Future research should explore hybrid models and real-time forecasting to enhance predictive accuracy and sustainability.