Optimization and comparative modelling of RSM and ANN for the adsorptive removal of Remazol Brilliant Blue R dye using spent coffee ground biochar

Abstract

The presence of dye pollutants in industrial wastewater poses serious environmental and health risks, necessitating efficient and sustainable treatment strategies. This study investigates the use of spent coffee ground biochar (SCGB), produced via low-temperature pyrolysis (350 °C), for the adsorptive removal of Remazol Brilliant Blue R dye. A Box–Behnken design with 27 experimental runs was employed to explore the influence of initial pH, adsorbent dosage, contact time, and initial dye concentration on dye removal efficiency. The coded values of the input variables were derived using standard transformation equations based on experimental ranges. Response surface methodology (RSM) and artificial neural networks (ANN) were developed and compared for modelling and optimization purposes. Under leave-one-out cross-validation (LOOCV), the best ANN with six hidden neurons achieved root mean square error (RMSE) = 5.1917 and coefficient of determination ($R^2$) = 0.9438, outperforming the RSM model (RMSE = 7.3587; $R^2$ = 0.8871). Using the full dataset, the ANN again showed higher accuracy ($R^2$ = 0.999; RMSE = 0.591) than RSM ($R^2$ = 0.973; RMSE = 3.630). The maximum experimental removal observed was 92.54 %. For process optimization within the experimental bounds, both models were optimized using a penalized objective to discourage unrealistically high predictions. RSM identified optima at 99 %, reflecting the steep rise of its quadratic surface at low pH, higher dosage, and longer time under the penalty. The ANN surface peaked near 95.4 %, showing smoother increases with diminishing gains in very favorable conditions. Overall, the ANN provides superior predictive accuracy, while RSM offers an interpretable baseline and suggests a higher theoretical maximum within the design space. Both models support a practical operating region characterized by low pH, higher adsorbent dosage, longer contact time, and a lower initial dye level when controllable. These findings highlight the promise of SCGB as a low-cost, sustainable adsorbent for dye-contaminated wastewater.