Biomass-Derived Activated Carbon for Congo Red Dye Adsorption: Machine-Learning-Based Prediction and Comparative Evaluation

Congo Red (CR) is a persistent carcinogenic azo dye frequently detected in industrial effluents, requiring reliable tools for predicting its removal performance under variable operating conditions. In this work, machine learning models were used to predict the adsorption behavior of activated carbon derived from Spathodea campanulata flowers (SCAC) for the removal of CR dye. A data set of 180 batch experiments (pH: 5–10, dosage: 0.2–1.0 g/L, initial CR: 20–60 mg/L, contact time: 0–180 min, temperature: 293–323 K) was used to construct and evaluate a set of four data-driven modeling approaches, namely support vector machine (SVM), adaptive neuro-fuzzy inference system (ANFIS), artificial neural network (ANN), and multiple linear regression (MLR). As a linear model, the interaction-type MLR model achieved a low R2 value (0.8667), indicating only partial capability to represent the nonlinear adsorption behavior. ANFIS showed the strongest overall fit among the tested models (R2 = 0.9722), closely followed by ANN (R2 = 0.9710). The medium-Gaussian SVM also showed strong predictive ability (R2 = 0.9597). Global sensitivity analysis using the optimized ANFIS model ranked the variables as contact time > initial concentration > pH > temperature > dosage. The results demonstrate that ANFIS offers a robust framework for accurately modeling and optimizing CR removal from wastewater by biomass-derived activated carbon.