Multilayer perceptron neural network-based QSAR models for the assessment and prediction of corrosion inhibition performances of ionic liquids

Abstract

The present study reports the quantum chemical studies and quantitative structure activity relationship (QSAR) modeling of thirty ionic liquids utilized as chemical additives to repress mild steel degradation in 1.0 M HCl. Five molecular descriptors obtained from standardization of calculated descriptors together with the inhibitor con-centration were employed in model building. Multiple linear regression (MLR) and multilayer perceptron neural network (MLPNN) modeling were utilized in model construction. The optimal MLPNN model was developed using a network architecture of 6-3-5-1 with Levenberg-Marquardt as the learning algorithm. The model yielded an MSE of 29.9242, RMSE of 5.4703, MAD of 4.9628, MAPE of 5.7809, rMBE of 0.1202 and CoV of 0.0052. The MLPNN model displayed better predictive performance than the MLR model. Furthermore, developed models were applied to forecast the inhibition efficiencies of five novel ionic liquids. The theoretical inhibitors were found to be effective inhibitors of steel corrosion, showing over 80% inhibition efficiency.