Machine learning-driven prediction of deep eutectic solvents’ heat capacity for sustainable process design

Heat capacity, a crucial physical property for chemical processes, is often understudied in Deep Eutectic Solvents (DESs), which in turn are promising green alternatives to environmentally hazardous conventional solvents. This work addresses this gap by developing a machine learning model to predict DES heat capacity and identify key structural features influencing it. We employed a dataset of 530 DESs with corresponding experimental heat capacity values. Quantum-chemical COSMO-RS-based descriptors, capturing detailed information about DES structures, were calculated for each data point. Various machine learning algorithms, namely k-Nearest Neighbours (kNN), Random Forests (RF), Neural Network Multilayer Perceptron (MLP), and Support Vector Machines (SVM) were explored alongside a linear model (Multiple Linear Regression, MLR). Hyperparameter optimisation ensured all models were fine-tuned for optimal performance. The most successful model, based on the MLP technique, achieved remarkably low Average Absolute Relative Deviation (AARD) values of 0.500 % and 3.999 % for the training and test sets, respectively. This signifies a significant improvement in prediction accuracy compared to traditional methods. Furthermore, by applying a SHapley Additive exPlanations (SHAP) analysis, we identified the most crucial structural factors within DES components that govern their heat capacity. This comprehensive investigation offers valuable insights that can pave the way for an efficient design of novel DESs in the future.