Document Type : Original Article
Authors
1
Department of Chemistry, PSG College of Arts and Science, Coimbatore, Tamil Nadu, India
2
Department of Information Technology, Gokaraju Rangaraju Institute of Engineering and Technology, Hyderabad, Telangana, India
3
Department of Chemistry, Dr. N.G.P. Institute of Technology, Anna University, Coimbatore-641048, Tamil Nadu, India
4
Department of Computer Science and Engineering, Vellore Institute of Technology, Chennai, Tamil Nadu, Indi
5
Higher Institute of Nursing Professions and Health Techniques of Agadir, Annex Guelmim, Agadir, Morocco
6
Laboratory of Spectroscopy, Molecular Modeling, Materials, Nanomaterials, Water and Environment LS3MN2E, CERNE2D, Faculty of Sciences, Mohammed V University in Rabat, Morocco
Abstract
Corrosion inhibitors are widely used to reduce metal degradation, although toxicity and experimental screening costs remain important concerns. In this study, a quantitative structure–property relationship (QSPR) machine-learning framework was evaluated for predicting the corrosion inhibition efficiency of quinoxaline derivatives for mild steel in 1 N sulfuric acid. The supplied modelling dataset contained 15 compounds and 11 density-functional-theory-derived quantum chemical descriptors. A descriptor-selection procedure based on simulated annealing was used to identify electronegativity (χ), LUMO energy (ELUMO), fraction of electrons transferred (ΔN), and energy gap (ΔE) as the most informative descriptor subset. Seven regression algorithms—linear regression, Lasso, Ridge, Elastic Net, decision tree, random forest, and gradient boosting—were compared. To address predictive validation explicitly, the data were divided reproducibly into 12 training compounds (80%) and 3 independent test compounds (20%) using random state = 42. All preprocessing and model fitting were performed using the training set, and final predictive statistics were calculated only on the held-out test set. The independent test results showed limited generalization because of the very small sample size; the lowest test MAE was obtained by linear regression (MAE = 4.48%, MSE = 21.14, R2 = −0.70). These results indicate that the selected electronic descriptors are mechanistically informative, but a larger chemical dataset is required before the QSPR models can be considered externally validated for routine prediction.
Keywords
Subjects
