Probe into the volumetric properties of binary mixtures: Essence of regression-based machine learning algorithms

Abstract

Evaluating the volumetric properties of the binary mixtures is always a cumbersome and tedious task as the volumetric properties are sensitive towards the environmental conditions. The present novel work aims to predict the volumetric properties of different binary mixtures using regression-based machine learning algorithms. Four different machine learning algorithms are employed for making the surrogate models, namely, Gradient Boosting Machine (GBM), Stacked Ensemble (SE), Random Forest (RF), and Extremely Randomized Trees (XRT) for predicting the density, ultrasonic speed, viscosity, deviation in ultrasonic speed, deviation in viscosity, and excess molar volume of the binary mix made up of p-chlorotoluene and methanol. The accuracy of the algorithms is evaluated using the statistical tools over the predicted and the actual values. The machine learning algorithms can successfully predict the volumetric properties of the mix with GBM as the best and XRT as the worst amongst the models studied. After that, the surrogate models are also built to predict the properties of the different binary mixtures, namely, 2-chlorotoluene and n-hexane, 4-chlorotoluene and n-hexane, 1,3-dichlorobenzene and n-hexane, 2-chlorotoluene and 1,4-dioxane, 4-chlorotoluene and 1,4-dioxane, 1,3-dichlorobenzene and 1,4-dioxane, 2-chlorotoluene and methanol, 2-chlorotoluene and ethanol, 2-chlorotoluene and cyclohexane, 4-chlorotoluene and cyclohexane, and 1,3-dichlorobenzene and cyclohexane. For all the binary mixtures studied, GBM is found to be the most accurate algorithm for predicting the properties. Density is found to be sensitive towards the temperature, whereas remaining properties towards the mole fraction.