Broad Learning System for Investigating Corrosion Inhibition Efficiency of Heterocyclic Compounds

Abstract

This study explores the use of Broad Learning Systems (BLS) to predict the corrosion inhibition efficiency (CIE) of heterocyclic compounds, addressing limitations of deep neural networks (DNNs) such as vanishing gradients and computational inefficiency. BLS prioritizes network width over depth, enabling faster learning and improved generalization. Trained on quantum chemical properties (QCPs) of 192 heterocyclic compounds, BLS outperformed multilayer perceptron neural networks (MLPNN) and random forest (RF) models, achieving lower mean absolute error (MAE: 1.41), root mean square error (RMSE: 1.79), and higher R² (0.993). Predicted CIE values for quinoxaline derivatives (95.39% and 94.05%) aligned closely with experimental data. This study demonstrates the potential of BLS as an efficient, accurate, and scalable approach for predicting corrosion inhibition capabilities, contributing to advancements in corrosion science and environmentally friendly solutions.   Keywords - machine learning, broad learning system, neural network, corrosion.