Tuned Machine Learning Modeling Tools and Genetic Algorithm for Anionic Dye Adsorption on Biomass-Based Activated Carbons

Abstract

In this study, the adsorption behaviors of Phragmites australis-based biochars and commercial activated carbon were compared for Telon Red dye removal. Phragmites australis-based biochars, C-RAC and HTC-RAC were produced by carbonization and hydrothermal carbonization methods, respectively. The adsorption capacities of CAC, C-RAC, and HTC-RAC were calculated as 262.66, 255.77, and 806.33 mg/g, respectively. Thermodynamically, Delta G degrees showed that the adsorption behavior for all adsorbents occurs spontaneously. Delta H degrees values showed that the dye adsorption on CAC and HTC-RAC were controlled by a physical mechanism, unlike C-RAC. The Delta S degrees values indicated that Telon Red molecules were randomly distributed on C-RAC during the adsorption processes, in contrast to CAC and HTC-RAC. Moreover, Gaussian Process Regression, Support Vector Regression, and Neural Network Regression with different strengths were used to model adsorption behaviors. From the optimal outcomes achieved through hyperparameter tuning, Gaussian Process Regression was the most effective modeling tool producing the closest-to-reality simulation. According to the MAPE criterion, an error rate of 0.8134 % for CRAC, 0.9324 % for CAC, and 0.7924 % for HTC-RAC were obtained in the simulations. The genetic algorithm was used to optimize these reliable and valid simulations to have the highest adsorption capacity with nearly desirability of 100 %. Finally, the adsorption capacities of all three adsorbents were compared statistically, and it was concluded that the adsorption capacity of HTC-RAC was significantly higher than the others.