Machine learning approach for mapping the heat capacity of deep eutectic solvents for sustainable energy applications

This study introduces a breakthrough machine learning (ML) approach to predict the heat capacity of deep eutectic solvents (DESs), an emerging class of environmentally friendly solvents, for sustainable energy applications. Leveraging an extensive dataset of 2,696 data points across 55 DES mixtures with 389 unique compositions, the research significantly extends the scope of previous studies. The methodology employs group contribution-based critical properties to elucidate intricate molecular-level dynamics, enabling the models to decode complex interactions governing DES behavior. The multi-dimensional strategy encompasses conventional binary DESs, ternary DESs, and water + DES mixtures, providing a holistic understanding of these green solvents. The developed artificial neural network (ANN) exhibits exceptional performance, with an R² value of 0.995 in training and 0.988 in testing, substantially surpassing existing models. Rigorous comparative analysis and applicability domain assessment confirm the robustness and superiority of this approach in predicting heat capacities for diverse DES compositions. Key findings include identifying the molecular factors influencing DES heat capacities, with critical volume, molecular weight, and critical temperature having the greatest impact. An open-source, user-friendly Excel tool based on the ANN model is developed to promote accessibility and practical application in energy and environmental sectors. This work revolutionizes the understanding and prediction of DES thermophysical properties, paving the way for their enhanced utilization in eco-friendly energy systems, thermal management, green materials development, and sustainable fuel processing.