Prediction and Interpretability Analysis of Deep Eutectic Solvent Viscosity based on a Stacking Ensemble Model

Abstract

Deep eutectic solvents (DES) are a new type of green solvent with significant application potential in the selective leaching of zinc-containing solid waste via hydrometallurgy. However, traditional design methods rely heavily on trial-and-error experimentation, resulting in low R&D efficiency, and the core physical property (viscosity) that determines mass transfer and reaction kinetics in the system is difficult to estimate accurately. To address this industry challenge, this paper proposes a high-precision viscosity prediction and micro-mechanism analysis framework that integrates multi-dimensional cross-feature engineering, the Whale Optimization Algorithm (WOA), and Stacking ensemble learning. Based on a rigorously cleaned, multi-source standardized dataset, the model employs Extreme Gradient Boosting (XGBoost) and Random Forest (RF)-both globally optimized via WOA-as base learners, combined with a Linear Regression (LR) meta-learner through a two-stage deep integration. Extrapolation evaluation on an independent test set demonstrates that this two-layer architecture effectively overcomes the generalization bottleneck in complex sample chemical spaces, achieving an excellent accuracy with a coefficient of determination ( ) of 0.8620 and an average absolute relative deviation (AARD) as low as 9.88%, with overall performance significantly outperforming various single-baseline models.Furthermore, this study introduced game-theoretic SHAP analysis, successfully breaking through the “black-box” barrier of deep ensemble models. The research quantitatively confirmed that the drastic changes in DES viscosity essentially stem from strong electrostatic coupling and a dense hydrogen-bond cross-linking network formed after component mixing.The positive synergistic interaction between high steric hindrance in the mixed space and strong electrostatic attraction constitutes the underlying mechanism driving the macroscopic viscosity jump, while the “coupled resonance” of multiple microscopic features at low temperatures serves as the fundamental driving force behind the sharp rise in hydrodynamic viscosity.This study not only provides a breakthrough data-driven paradigm for estimating the physical properties of complex chemical systems, but also offers solid quantitative theoretical guidance for the targeted reverse design of low-viscosity, high-performance green DES solvents for industrial zinc extraction through a multidimensional, mechanism-transparent analysis.