Machine learning-driven optimizations of phosphate removal using agricultural waste-based adsorbents: integrating RFR–RSM, ANN–CCD, GPR–BBD, and SVR–TOA models for enhanced adsorbent efficiency with agro-based adsorbents

Phosphate contamination caused by eutrophication continues to degrade aquatic ecosystems. This study developed agro-biomass adsorbents from rice husk and coconut shell and optimized phosphate removal using integrated machine learning and experimental design approaches. Random forest–RSM, ANN–CCD, GPR–BBD, and SVR–Taguchi models were applied to examine nonlinear effects of pH (3–9), temperature (20–60 °C), adsorbent dosage (0.5–5 g/L), and initial phosphate concentration (10–100 mg/L). Under optimized conditions, the integrated model achieved 93% phosphate removal and 50 mg/g adsorption capacity. ANN–CCD produced the highest adsorption capacity, while GPR–BBD quantified prediction uncertainty within ±2%. Compared with conventional optimization, removal efficiency increased by 6%–8%, and adsorption capacity improved by 7–10 mg/g. The proposed framework reduces experimental runs and supports scalable, sustainable phosphate removal using agro-based adsorbents.