Machine learning-powered qualitative structure properties relationship models for prediction of corrosion inhibition efficiencies of triazoles.

The development of a holistic theoretical framework that can predict the corrosion inhibition efficiency and evaluate the anti-corrosion potentials of novel materials has been a challenging one. This work aimed to address this challenge by integrating machine learning-based Quantitative Structure-Property Relationship (QSPR) models with computational simulation techniques. 25 descriptors derived from density functional theory (DFT) results for 130 triazole derivatives on mild and carbon steels in hydrochloric acid (HCl) solutions were used to develop predictive models. Random Forest, K-Nearest Neighbor, Gradient Boosting, Support Vector Regression, and Stacked Regression. Key features influencing the model's predicted outcomes were identified through Recursive Feature Elimination (RFE) and Shapley Additive ExPlanation (SHAP) analyses. The models demonstrated competitive performance with Stacked Regression being more pronounced, as indicated by results of some statistical metrics, including Mean Squared Error (60.50-64.90), Root Mean Squared Error (7.740-8.056), Mean Absolute Error (6.179-6.340), Mean Absolute Percentage Error (7.17-7.37), and a concordance correlation coefficient (0.30-0.31). Among the novel triazoles evaluated, T