Adsorption simulation of 2,4-D pesticide on novel zinc-based 2-amino-4-(1H-1,2,4-triazole-4-yl)benzoic acid coordination complexes using machine learning approach.

The capacity of zinc-based 2-amino-4-(1H-1,2,4-triazole-4-yl)benzoic acid coordination complex (Zn(NH-TBA)) and modified Zn(NH-TBA)COMe complex for removal of 2,4-dichlorophenoxyacetic acid (2,4-D) from aqueous solutions was investigated through adsorption modeling and artificial intelligence tools. Analyzing the adsorption characteristics of pesticides helps in studying the groundwater pollution by pesticides in agriculture area.In this study, Zn(NH-TBA) was synthesized using Schiff base and its surface was modified using acetic anhydride group and their physical characteristics were identified using proton NMR, FTIR, and XRD. NMR results showed maximum modification yield obtained was 65% after 5 days. The porous structure and surface area monitored using nitrogen isotherm and BET surface area analysis presented relatively less surface area and porosity after modification. Adsorption modelling indicated that Toth model with a maximum adsorption capacity of 150.8 mg/g and 100.7 mg/g represents the homogenous adsorption systems which satisfy both low- and high-end boundary of adsorbate concentration in all settings according to the optimum point, while the kinetics and rate of 2,4-D adsorption follow the pseudo-first-order kinetic model in all situations. Artificial neural network (ANN), support vector regression, and particle swarm optimized least squares-support vector regression (PSO-LSSVR) were used for the optimization and modelling of adsorbent mass, adsorbate concentration, contact time, and temperature to develop predictive equations for the simulation of the adsorption efficiency of 2,4-D pesticide. The obtained results exhibited the better performance of ANN and PSO-LSSVR for prediction of adsorption results. The mean square error values of ANN (0.001, 0.012) and PSO-LSSVR (0.121, 0.105) were obtained for Zn(NH-TBA) and Zn(NH-TBA)COMe, respectively, while their respective coefficient of determination (R) obtained were 0.999 and 0.988 for ANN and 0.980 and 0.825 for PSO-LSSVR. The study specified that machine learning predictive behavior performed better for Zn(NH-TBA) compared to Zn(NH-TBA)COMe that is also supported by theoretical kinetics and isotherm models. The research concludes that artificial intelligence models are the most efficient tools for studying the predictive behavior of adsorption data.