Multivariate analysis and soft computing-based prediction of energy potential in heterogenous waste streams

This study presents a data-driven framework for characterizing waste-derived biomass for energy recovery. Utilizing a dataset comprising higher heating value (HHV), elemental composition, and proximate properties of diverse waste streams, correlation analysis and feature importance analysis (FIA) using Random Forest (RF)’s importance metrics were conducted to identify key parameters influencing HHV prediction. Carbon and and Hydrogen were identified as the most significant contributors, accounting for 75–80% of the model’s predictive strength. Principal Component Analysis (PCA) was applied to cluster waste types based on compositional and energetic similarities, aiding in the classification of waste for optimized waste-to-energy (WtE) strategies. Dimensionality was effectively reduced with over 90-95% of variance captured in the first four principal components. The predictive performance of three machine learning models—Artificial Neural Network (ANN), Support Vector Machine (SVM). The RF model demonstrated superior performance during training with RMSE, MAE, MAD, and rMBE values of 0.8606, 0.5945, 0.3864, and 0.0895, respectively. This integration of statistical techniques and machine learning provides a robust tool for waste classification and HHV estimation, promoting data-informed decisions in sustainable waste management and energy.