Advanced machine-learning frameworks for predicting malachite green adsorption onto superparamagnetic-activated carbon derived from Spathodea campanulata flowers

This investigation introduces a comprehensive machine-learning (ML) scheme for accurately predicting the uptake of the cationic dye Malachite Green (MG) onto superparamagnetic activated carbon derived from Spathodea campanulata flowers. Batch adsorption experiments data covering a wide range of solution pH, sorbent dosage, initial dye concentration, contact time, and temperature were used to compare four predictive models: multiple linear regression (MLR), artificial neural network (ANN), adaptive neuro-fuzzy inference system (ANFIS), and support vector machine (SVM). Pearson correlation analysis revealed contact time as the dominant positive factor influencing adsorption capacity (Q_e). Among MLR variants, the interaction-linear model achieved the best fit (R² = 0.8901). SVM with a medium Gaussian kernel improved accuracy substantially (R²= 0.9577). ANN delivered similarly strong predictive power (overall R² =0.9672) by learning complex multidimensional patterns. ANFIS emerged as the most robust and generalizable model, achieving R² = 0.9683 with the lowest mean squared error (0.0098), root mean squared error (0.0992), and mean absolute error (0.0311). Sensitivity analysis of the optimized ANFIS confirmed the primacy of contact time (56.8 %), followed by initial concentration (33.2 %), dosage (3.6 %), temperature (3.3 %), and pH (3.1 %). This integrated experimental–computational approach offers a scalable, data-driven strategy for designing magnetically recoverable adsorbents and optimizing dye remediation in complex wastewater matrices.