Comparative evaluation of artificial intelligence based prediction of dye removal by ultrasonic activated carbon composites derived from mixed biomass: Optimization and sensitivity analysis using Garson and Pertubation method

Abstract

Development of efficient and sustainable adsorbent along with predictive tools has become crucial for the effective pollutant remediation strategies. The current research introduces a novel integration of ultrasonication modified algal-manila tamarind seed biomass (UAMTB) with Artificial intelligence based predictive modelling for the effective dye removal. The potential surface and functional nature has been analysed by material characterization studies. Highest adsorption capacity of 298.85 mg/g for Basic Orange (BO) and 283.88 mg/g for Eriochrome Black (EB) dye has been observed at UAMTB dose of 1.2 g/L within a time interval of 25 min. In the AI model training comparing Artificial Neural Network (ANN) and Adaptive Neuro Fuzzy Inference System (ANFIS), best performance was obtained for the tansig purelin model with optimal neural architectures of [4 25 1] for BO dye and [4 30 1] for EB dye with higher R value of 0.9977 and 0.9986 for EB and BO dye respectively in ANN model. The Mean squared Error (MSE) value for the tansig-purelin model was found to be 0.0012 with ANN model which was higher than ANFIS MSE value of 1.13. Sensitivity analysis using Garson and Perturbation method observed that the UAMTB dose is the most influential input variable in the current adsorption process. The dual model approach of sensitivity analysis validates the efficacy of ANN modelling for process optimization in complex adsorption systems.