Insights into corrosion inhibition development through QSAR and machine learning: Application to benzimidazole derivatives

• H. El-Idrissi^1,2, A. Diane¹, M. Driouch², M. Lahyaoui¹, N. Saffaj³, R. Mamouni³, B. Ihssane^1,4, T. Saffaj¹, A. Haoudi¹, A. Mazzah⁵, M. Sfaira² and A. Zarrouk⁶

¹ Laboratory of Applied Organic Chemistry, Sidi Mohamed Ben Abdellah University (USMBA), Faculty of Sciences and Techniques, Route Imouzzer P.O. Box 2626 Fez, Morocco
² Laboratory of Engineering, Modeling and Systems Analysis (LIMAS). Sidi Mohamed Ben Abdellah University (USMBA), Faculty of Sciences, P.O. Box 1796-30000, Fez-Atlas, Morocco
³ Laboratory of Materials and Biotechnology and Environment (LBME), Faculty of Sciences, Ibn Zohr University, Agadir, Morocco
⁴ Physio-Chemical Laboratory of Inorganic and Organic Materials (LPCMIO), Ecole Normale Supérieure, Materials Science Center (MSC), Mohammed V University, Rabat, Morocco
⁵ University of Lille, CNRS, USR 3290, MSAP, Miniaturization for Synthesis, Analysis and Proteomics, Lille, France
⁶ Laboratory of Materials, Nanotechnology and Environment, Faculty of Sciences, Mohammed V University in Rabat, Av. Ibn Battuta. P.O. Box 1014, Rabat, Morocco

Abstract: As corrosion inhibitors for mild steel in 1 M hydrochloric acid, benzimidazole derivatives were studied by applying the quantitative structure-activity relationships (QSAR). A total of nine models were built based on selected variables, and a further nine models were built using 2D descriptors derived by Molecular Operating Environment (MOE) software. The models Multiple linear regression (MLR), Support Vector Machine Regression (SVR), Back Propagation Artificial Neural Networks (BPANN), Partial Least Square (PLS), and others were created to forecast the inhibitory effectiveness of the new benzimidazole compounds. In addition, the predictive model’s calibration and test results’ coefficients of determination (R²) and root mean squared (RMSE) were evaluated. Despite most models showing adequate performance, the SVR model was chosen as the best model due to its successful calibration and test results represented in higher coefficient of determination of calibration and test, which are 97.83% and 94.54%, respectively, and the lower RMSE of calibration and test which are 2.45 and 2.96. Furthermore, plotting the anticipated inhibition efficiency versus the experimental value revealed a good degree of dependability for the predictions made by the SVR model for the novel inhibitors. The selection of the descriptors and their effects on the effectiveness of inhibition were essential factors in creating and developing new benzimidazole derivatives as corrosion inhibitors. The SVR model allows for precise prediction of the inhibitory efficacy of new benzimidazole derivatives. This QSAR model offers helpful insights into developing and optimizing novel benzimidazole derivatives as corrosion inhibitors, emphasizing the significance of descriptor selection and the impact of each descriptor on the efficacy of inhibition. Therefore, this study demonstrates how using Deep Learning and Machine Learning techniques in QSAR analysis can improve the accuracy and reliability of the outcomes.

Keywords: QSAR modeling, corrosion inhibition, meta-heuristic optimization, machine learning, deep learning

Int. J. Corros. Scale Inhib., 2023, 12, no. 4, 2101-2128
doi: 10.17675/2305-6894-2023-12-4-36

Download PDF (Total downloads: 517)

Back to this issue content: 2023, Vol. 12, Issue 4 (pp. 1365-2495)