Machine Learning to Predict Melting Temperatures of Deep Eutectic Solvents from Molecular Descriptors

Deep Eutectic Solvents (DESs) emerged as promising alternatives to conventional solvents for their peculiar characteristics. Among these, melting temperature is a critical parameter that deeply influences their industrial applicability. It is possible to use machine learning (ML) models to predict the melting temperatures of DES mixtures based on their molecular features, to reduce time-consuming and costly experimental work. The approach involves training and testing a series of existing learning models, using the code developed by a separate research group, on an expanded dataset curated by a colleague. Moreover the possibility of modifying the molecular descriptors originally used as input for prediction will be explored. The impact of the changes implemented in the original code will be assessed using statistical metrics, which also allows to compare each model's performance. The results should provide valuable insight into the influence of feature selection and dataset size on prediction accuracy, and confirm the most effective model for melting temperature prediction.