Gravimetric inhibition efficiency prediction model of AA7075-T7351 alloy using Treculia africana extract in 1.0 M HCl through input feature optimization

Abstract

The applications of four machine learning (ML) algorithms, namely: Support Vector Regressor (SVR), Extreme Gradient Boosting (XGBoost), Least absolute shrinkage and selection operator (Lasso), and Ridge, in predicting the corrosion inhibition efficiency (IE) of Treculia africana (TA) leaves extract on AA7075-T7351 alloy, in corrosive 1.0 M HCl environment, with a small (42) sample space, have been studied. Time and resource constraints in traditional corrosion study methods have been avoided through feature engineering to expedite prediction process. The dominant features, which affected the IE, were done through feature importance and selection processes using pair plot matrix of features and Kendall correlation etc., to remove redundant features. The results in the form of data visualization, feature importance, and the performance of each algorithm on the test set were explicitly depicted. The evaluation metrics, including coefficients of determination (R2) and root mean square error (RMSE), validated the efficacy of the models in predicting the IE of TA on AA7075-T7351 in 1.0 M HCl environments. Ridge model demonstrated superior accuracy, with R2 score of 0.972, particularly in handling the highly correlated dataset used in this study. SVR followed closely in performance (0.969). XGBoost proved reliable at R2 score of 0.953. Lasso with R2 of 0.952 was the least of the four models, due to its random feature selection method. The RMSE scores corroborated the prediction accuracies with values; 4.145, 4.408, 5.138 and 5.462 respectively. This study revealed the viability of using the four machine learning algorithms in potential generalization ability of IE prediction accuracy, while offering an efficient and accurate alternative to traditional methods. Graphical Abstract