This study investigated the arsenide removal by using mesoporous CoFe2O4/ graphene oxide nanocomposites based on batch experiments optimized by artificial intelligence tools. These nanocomposites were prepared by immobilizing cobalt ferrite on graphene oxide and then characterized using various techniques, including small angle X-ray diffraction, high-resolution transmission electron microscopy and energy-dispersive X-ray spectroscopy. Artificial intelligence tools associated with response surface methodology were employed to optimize the conditions of the arsenide removal process. The results showed that back propagation neural network combined with genetic algorithm was suitable for the arsenide removal from aqueous solutions by the nanocomposites based on the minimum average values of absolute errors and the value of R2. The optimal values of the four variables (operating temperature, initial pH, initial arsenide concentration, and contact time) were found to be 25.66 °C, 7.58, 10.78 mg/L and 46.41 min, and the predicted arsenide removal percentage was 84.78%. The verification experiment showed that the arsenide removal percentage was 86.62%, which was close to the predicted value. Three evaluation methods (gradient boosted regression trees, Garson method and analysis of variance) all demonstrated that the temperature was the most important explanatory variable for the arsenide removal. In addition, the arsenide removal process can be depicted with pseudo-second-order kinetics model and Langmuir isotherm, respectively. The thermodynamics investigation disclosed that the adsorption process was of a spontaneously endothermic nature. In summary, this study showed that ANN-GA was an efficient and feasible method in determining the optimum conditions for arsenic removal by CoFe2O4/ graphene oxide nanocomposites.

    Abstract
    1 Introduction
    2 Materials and methods
        2.1 Materials
        2.2 Preparation and characterization of graphene oxide and CoFe2O4GO nanocomposites
        2.3 Batch adsorption tests
        2.4 RSM based modeling
        2.5 ANNs modeling
        2.6 Random forest for optimization
        2.7 Radial basis function neural network
    3 Results and discussion
        3.1 Characterization of CoFe2O4GO nanocomposites
        3.2 CCD for optimizing the process parameters of arsenide removal
        3.3 The modeling and optimizations of AI tools
        3.4 Comparison among analysis of variance, GBRT and Garson method
        3.5 The characteristics of arsenide adsorbed on Co2Fe2O4GO
        3.6 Analysis of adsorption mechanism
    4 Conclusion
    Acknowledgements 
    References

• Xianliang Wu(Guizhou Provincial Key Laboratory for Information Systems of Mountainous Areas and Protection of Ecological Environment, Guizhou Normal University, Guizhou Institute of Biology)
• Jiwei Hu(Guizhou Provincial Key Laboratory for Information Systems of Mountainous Areas and Protection of Ecological Environment, Guizhou Normal University, Cultivation Base of Guizhou National Key Laboratory of Mountainous Karst Eco‑Environment, Institute of Karst, Guizhou Normal University)
• Xingfu Wang(Cultivation Base of Guizhou National Key Laboratory of Mountainous Karst Eco‑Environment, Institute of Karst, Guizhou Normal University)
• Ling Xin(Guizhou Provincial Key Laboratory for Information Systems of Mountainous Areas and Protection of Ecological Environment, Guizhou Normal University)
• Caifang Li(Guizhou Provincial Key Laboratory for Information Systems of Mountainous Areas and Protection of Ecological Environment, Guizhou Normal University)
• Xionghui Wei(Department of Applied Chemistry, College of Chemistry and Molecular Engineering, Peking University)