Novel Predictive Models for the Heat Capacity of Deep Eutectic Solvents Using Coupled Atomic/Group Contributions and Machine Learning Methods

Abstract

Deep eutectic solvents (DESs) are novel green solvents. Potential applications of DESs require a knowledge of their physical and thermodynamic properties. This study is devoted to the DES heat capacity. Since the potential number of DESs to be prepared in the future is innumerable, it is vital to have predictive models. In this study, two machine learning models, namely, the multilayer perceptron artificial neural network (MLPANN) and the least square support vector machine (LSSVM) were coupled with the group contribution (GC) and atomic contribution (AC) approaches. In the contribution methods, each structural fragment of the compounds is considered as input to the machine learning models, significantly enhancing predictive capability. A comprehensive database was collected, including 640 data points from 51 different DESs at various temperatures. The MLPANN-GC and LSSVM-GC models resulted in AARD% values of 1.74 and 1.73%, respectively, while the corresponding values were 2.90 and 2.64% for the MLPANN-AC and LSSVM-AC models.