Alginate (ALG) is an important biomaterial, with particular properties, which showed its utility for various applications in chemical and environmental engineering, biomedical science, and other industries (e.g. food, packing, paper, and textile) [1]. We report herein a machine learning approach based on an artificial neural network (ANN) to estimate the adsorption performance of a composite adsorbent with magnetic properties. This magsorbent was formulated as a cross-linked alginate matrix with glutaraldehyde (ALG-GA) that incorporated 15% wt. inorganic nanoparticles of cobalt ferrite (CoFe2O4). The produced magsorbent (CoFe2O4@ALG-GA) was well characterized by instrumental physical-chemical techniques (FTIR, VSM, SEM, and EDX). Afterward, this magnetic material was utilized for the adsorption of a persistent organic pollutant (Rhodamine, Rh6G) from synthetic wastewaters. Adsorption assays were done to evaluate the kinetics, isotherms, and thermodynamics parameters. According to Dubinin-Radushkevich isotherm, the computed free energy of adsorption (ES) varied from 9.4 to 10.6 (kJ/mol) revealing an ion exchange as the predominant retention mechanism [2]. The adequate prediction capability of the trained ANN-model allowed determining the optimal adsorption conditions by using model-based optimization. Under optimal conditions established, the maximal removal efficiency of 96.54% was confirmed experimentally for an initial concentration of 110 mg/L of the Rh6G pollutant. In addition, the details about the interactions at the molecular level were estimated by the molecular modeling methods. Hence, the molecular docking results suggested that the binding of the Rh6G cationic dye (ligand) to the cross-linked alginate matrix (receptor) was based on the hydrophobic and electrostatic interactions. Moreover, the molecular dynamics simulation highlighted the stability grade of the docked complex versus the simulation time (Fig.1). Fig.1. Molecular dynamics results showing the interactions between ALG-GA (receptor) and Rh6G (ligand).Acknowledgment. This work was supported by a grant of the Romanian Ministry of Research, Innovation and Digitization, CNCS/CCCDI – UEFISCDI, project number PN-III-P1-1.1-TE-2019-0594, within PNCDI III.