Application of Machine Learning Methods To Removal Percentage Prediction for Malachite Green Adsorption on Kaolinite

Abstract

In this study, the removal percentage was estimated using machine learning methods, such as artificial neural network, radial basis function neural network, support vector regressor, and random forest regressors, for data obtained during Malachite green adsorption on kaolinite as an adsorbent in an aqueous solution. Important process parameters, including initial dye concentration, sonication time and temperature, were investigated. Statistical evaluation metrics such as R2, mean squared error, and root mean square error were used to evaluate the performance of the models. Among these models, the artificial neural network was more successful compared to other models with 0.98 R2 values for three temperatures. Radial basis function neural network and random forest regressors were observed to achieve successful results. In this study, the results obtained from the machine learning methods are given comparatively. The initial dye concentrations increased from 10 to 60 mg L-1, the removal percentage of Malachite green on kaolinite increased from 68.71% to 79.61% for 298 K, 72.26% to 82.58% for 308 K and 78.75% to 85.91% for 318 K, respectively. Isotherm, kinetic and thermodynamic calculations for Malachite green removal by kaolinite were completed. The equilibrium of Malachite green adsorption onto kaolinite was best described by the Langmuir isotherm and the kinetics of the process followed the pseudo-second-order model, which had the highest correlation values. Thermodynamic analysis of experimental data suggests that the adsorption process is spontaneous and endothermic in nature.