Insights into levofloxacin adsorption with machine learning models using nano-composite hydrochars.

Hydrothermal carbonization was applied to taro peel wastes to produce hydrochars using a facile and environmentally friendly process. Four different entities were prepared: hydrochar (TPh), phosphoric-activated hydrochar (P-TPh), and silver@hydrochars (Ag@TPh, Ag@P-TPh). The elemental compositions of the single and composite hydrochars were confirmed by EDX. Among the produced hydrochars, the morphology of the Ag@hydrochar composites demonstrated more wrinkled structure, and Ag nanoparticles decorated the surface. The optimal experimental conditions for levofloxacin adsorption were determined to be a contact time of 45 min, hydrochar dose of 0.15 g L, and pH of 7. The best adsorption performances were assigned to Ag@hydrochars. Two machine learning models were applied to predict the levofloxacin adsorption efficiency of the Ag@hydrochars. A central composite design (CCD) and a 3-10-1 artificial neural network (ANN) model were developed to estimate the removal performance of levofloxacin using Levenberg-Marquardt backpropagation algorithm based on correlation and error analysis of the adopted training functions. Furthermore, the ANN sensitivity analysis revealed the order of the relative importance variable as initial concentration> hydrochar dose> pH. The predicted values of the CCD and ANN models fitted the experimental results with R> 0.989. Therefore, the applied models were effective in predicting levofloxacin removal under different operating conditions. This work provides an open option for the sustainable management of food industry wastes and the possibility of waste valorization to effective hydrochar composites to be applied in water treatment processes.