MODELING AND PREDICTION OF THE THERMAL CONDUCTIVITY OF IONIC LIQUIDS USING FLAML AS AN AUTOMATED MACHINE LEARNING APPROACH

This manuscript reports the development of models using Microsoft’s library FLAML as automated machine learning approach to predict the behavior of thermal conductivity for ionic liquids. The models were obtained using an automated machine learning approach, the group contribution principle and Kernel Ridge Regression. A total of 990 experimental thermal conductivity data points for 81 ionic liquids were used to develop and compare the performance of 3 models. The results showed that the tested models calculated the thermal conductivity of ionic liquids with MAPE values lower than 2% for used database. The model trained with a split of 60/40 for training and test datasets respectively, showed the best performance for the prediction of the thermal conductivity of these compounds which points out at a good generalization capacity.