Prediction of molecular diffusivity of organic molecules based on group contribution with tree optimization and SVM models

Abstract

Prediction of mass transfer diffusion coefficient as a functional of group contribution in the molecular structure was carried out using machine learning techniques. Two machine learning methods including Tree Optimization (TO) and SVM (Support Vector Machine) were implemented to simulate the values of diffusion coefficients for various compounds. A bunch of data were collected from references for the diffusivity of nonelectrolyte organic molecules at infinite dilution in aqueous solution. The results can be beneficial for design and applications for wastewater treatment processes where the organic molecules must be removed from aqueous streams. For the modeling, 148 diverse functional groups were taken into account as the model's inputs, while the diffusivity of the compound was taken as the sole model's output in the computational study in this work. For modeling of the diffusion coefficients, 3000 datasets are chosen at random for the training procedure of the machine learning models. The simulation results revealed that the optimized Tree model is better at estimating the output parameter. The SVM model, on the other hand, can only forecast the outcome marginally with low accuracy compared to the Tree model.