ABSTRACT Recent experimental studies suggest that various heterocyclic organic molecules have good anticorrosion properties on mild steel in aqueous HCl medium. However, screening a large number of such molecules for anticorrosion activities is experimentally challenging. To address this challenge, we have applied a Quantum Mechanics based Machine Learning (ML) approach on a set of benzothiazole (BT) derivatives to predict new molecules that might have better corrosion inhibition efficiency. To overcome the small number of BT derivatives with experimentally available data on corrosion inhibition, theoretical calculations using high‐level density functional theory and molecular dynamics simulations on structurally similar molecules with various electron‐donating and electron‐withdrawing groups were performed and added to the dataset. Using data augmentation, we first expand the dataset in order to reduce the biasness of the dataset used in different ML models. Using correlation matrix, we find that few properties such as dipole moment, Molecular surface area, energy of the Frontier molecular orbitals etc. are highly correlated and contribute maximum to the adsorption energy property. We feed these data to three distinct ML models (Tabular Neural Network, Light Gradient Boosting Method, and Random Forest Regressor) and predict new BT derivatives that might have better corrosion inhibition properties without performing experiments.
Rout et al. (Fri,) studied this question.