Optimization and modeling of methyl orange adsorption onto polyaniline nano-adsorbent through response surface methodology and differential evolution embedded neural network.

Presence of pigments and dyes in water bodies are growing tremendously and pose as toxic materials and have severe health effects on human and aquatic creatures. Treatments methods for removal of these toxic dyes along with other pollutants are growing in different dimensions, among which adsorption was found a cheaper and efficient method. In this study, the performance of polyaniline-based nano-adsorbent for removal of methyl orange (MO) dye from wastewater in a batch adsorption process is studied. Along with this to minimize the number of experiments and obtain optimal conditions, a multivariate predictive model based on response surface methodology (RSM) is developed. This is compared with data-driven modeling using the artificial neural network (ANN) which is integrated with differential evolution optimization (DEO) for prediction of the adsorption of MO. The interactive effects on MO removal efficiency with respect to independent process variables were investigated. The fit of the predictive model was found to good enough with R2 = 0.8635. The optimal ANN architecture with 5-12-1 topology resulted in higher R2 and lower RMSE of 0.9475 and 0.1294 respectively. Pearson's Chi-square measure which provides a good measurement scale for weighing the goodness of fit is found to be 0.005 and 0.038 for RSM and ANN-DEO respectively, and other statistical metrics evaluated in this study further confirms that the ANN-DEO is very superior over RSM for model predictions.