Walnut Shell–Rice Husk Hybrid Adsorbent for Efficient Phenol Removal: Characterization, Kinetics, and Neural Network Prediction

• Walnut Shell–Rice Husk hybrid adsorbent was developed and characterized for adsorption experiments. • Adsorption isotherm models and kinetic analyses were conducted to evaluate adsorption behaviour. • The surface morphology and crystalline structure of the adsorbent were examined using SEM and XRD techniques. • Process optimization was performed using SN ratio and ANOVA , predict the experimental results by LM algorithm This research introduced Walnut Shell- Rice Husk hybrid adsorbent synthesized by combining Walnut Shell and Rice Husk powders to enhance the removal of phenol from the wastewater. Nine adsorption experiments were conducted by systematically varying four critical process parameters. Among the three isotherm models, the Freundlich model provided the best correlation (R² = 0.9979), suggesting a favourable multilayer adsorption process on a heterogeneous surface, outperforming both Langmuir (R² = 0.9826) and Temkin (R² = 0.9923) models. Kinetic analysis indicated pseudo-second-order model provided the superior fit, with theoretical equilibrium capacities closely matching experimental data when compared to the first-order model. Scanning Electron Microscopy analysis revealed the heterogeneous, porous Walnut Shell–Rice Husk structure with silica flakes and fibrous roughness. The X-ray Diffraction (XRD) pattern exhibited several well-defined and sharp diffraction peaks at 2θ values that match JCPDS code 46-1045, confirming α-quartz (SiO₂) with distinct crystallographic planes. From the Optimization process, the highest adsorption capacity was achieved at 60 mg/L phenol concentration, 6 g adsorbent dosage, 400 °C Pretreatment temperature, and 243 min of contact time and optimal phenol removal efficiency occurred at a lower concentration of 40 mg/L with a higher adsorbent dosage of 9 g under 400 °C Pretreatment temperature conditions. Additionally, an Artificial Neural Network (ANN) effectively predicted the results by learning from experimental data related to output responses. The ANN model accuracy was assessed using Mean Squared Error (MSE), Error histograms, and Regression fit plots, which help to evaluate how well the predicted outputs match the actual values.