Can Machine Learning Predict the Phase Behavior of Surfactants?

We explore the prediction of surfactant phase behavior using state-of-the-art machine learning methods, using a data set for twenty-three nonionic surfactants. Most machine learning classifiers we tested are capable of filling in missing data in a partially complete data set. However, strong data bias and a lack of chemical space information generally lead to poorer results for entire de novo phase diagram prediction. Although some machine learning classifiers perform better than others, these observations are largely robust to the particular choice of algorithm. Finally, we explore how de novo phase diagram prediction can be improved by the inclusion of observations from state points sampled by an analogy to commonly used experimental protocols. Our results indicate what factors should be considered when preparing for machine learning prediction of surfactant phase behavior in future studies.

Automated summary

We used advanced machine learning methods to explore the prediction of surfactant phase behavior, using a data set for twenty-three nonionic surfactants. Most machine learning classifiers tested were able to fill in missing data in a partially complete data set. However, strong data bias and a lack of chemical space information generally resulted in poorer results for entire de novo phase diagram prediction. Although some machine learning classifiers performed better than others, these observations were largely robust to the particular choice of algorithm. Finally, we examined how de novo phase diagram prediction can be improved by including observations from state points sampled by an analogy to commonly used experimental protocols. Our findings indicate the factors that should be considered when preparing for machine learning prediction of surfactant phase behavior in future studies.