Enhanced bio-adsorbent derived from turnip leaves for crystal violet removal in aqueous solutions: experimental investigation, characterization, and machine learning modeling

Abstract The consumption of vegetables generates a lot of waste, allowing their use as biomass. biosorbent based on turnip leaves were prepared; they were processed in the raw state (TL) and activated with H 3 PO 4 (TLA), and then tested to remove the Crystal Violet dye CV from aqueous solutions. Adsorbents were characterized using (FTIR), (SEM), (XRD), (TGA) and pH PZC , revealing that the functional groups OH, C-H, C=O and C-O-C were mainly responsible for the adsorption of CV. Scanning electron microscope (SEM) imaging revealed cellulose fibers with a multicellular structure initially linked to lignin and hemicellulose, which dissociated after chemical treatment, XRD analysis confirmed the amorphous nature of the structure, attributed to the presence of lignin and hemicellulose. The kinetic study showed that the best suited models to describe the experimental data were the pseudo-second-order (PSO) model for TLA and TL. The analysis of the adsorption isotherms using different models showed that they followed the Sips isotherm with a maximum adsorption capacity (qmax) of 635.54 mg/g for TL and 621.76 mg/g for TLA. The optimal conditions were found to be an adsorbent mass of 20 mg, a temperature of 25 °C. and a concentration of 10 mg/L with respective contact times of 150 min for TL and 120 min for TLA, leading to elimination yields of 92.60 % and 97.56 %. The mechanism of adsorption of the CV dye using TL and TLA could be explained by electrostatic interactions between the negatively charged surface of TL and TLA and the positively charged group of the CV dye. A thermodynamic study was carried out on the process of adsorption of the CV dye from aqueous solutions using activated and raw turnip leaves which revealed that this process was exothermic and spontaneous due to the approximate values of ΔH (−25.26 for TL and −20.69 for TLA) and ΔG. The predictive capacity of the multi-component adsorption system was studied using a Support Vector Machine (SVM) model. Two SVM approaches were developed and compared. The first model involved the use of the SVM method with an integrated optimization algorithm, while the second model used a more recent and efficient method, the Dragonfly (DA) algorithm, in conjunction with the SVM method. The evaluation of the accuracy of the models was carried out using three commonly used statistical measures: the mean squared error (RMSE), the coefficient of determination (R2) and the correlation coefficient (R). The data included five important variables, from 136 observations: the weight of the turnip leaves (TL or TLA), the initial concentration (c 0 in mg/L), the dose of foliar waste (in mg/L), the time (in minutes) and the final concentration (Cf). The programming was carried out using the MATLAB software. The results demonstrated that the DA-SVM model optimized with an RBF-Gaussian kernel function had an excellent prediction ability, with an R2 of 0.997, an R of 0.998 and an RMSE of 1.0809.