Machine learning-assisted evaluation of pH and scavenger effects on methylene blue photodegradation using green-synthesized Nb2O5 nanoparticles

The expansion of industrial activities has significantly increased water contamination by organic pollutants, especially persistent synthetic dyes (e.g., methylene blue - MB), which pose environmental and health risks. In this context, sustainable and efficient treatment technologies are required. Heterogeneous photocatalysis has emerged as a promising approach, particularly when combined with green-synthesized nanomaterials. In this study, niobium pentoxide nanoparticles (Nb 2 O 5 -NPs) were synthesized via a green route using Carya illinoinensis nutshell extract and applied in the photodegradation of MB under visible light, with emphasis on pH effect. The nanomaterials were characterized by X-ray diffraction (XRD), zeta potential (ZP), zero charge point (pH ZCP ), and diffuse reflectance spectroscopy (DRS). Additionally, Machine Learning (ML) models (Decision Tree, Random Forest, and Bagging) were employed to evaluate the influence of reactive species scavengers. The Nb 2 O 5 -NPs exhibited a pseudohexagonal crystalline structure, average crystallite size of 21.6 ± 8.1 nm, surface charge of −22.7 ± 1.7 mV, pH ZPC of 6.77, and band gap energy of 3.0 eV. Photocatalytic experimental results showed that pH strongly influences MB degradation, with photodegradation increasing from 16.0% to 60.4% and kinetic constants rising from 0.0016 min −1 to 0.0059 min −1 . Among the ML models, Bagging presented the best predictive performance (R 2 training = 0.90; R 2 test = 0.83), confirming that OH and O 2 − radicals are the main contributors. Feature importance analysis revealed that reaction time, pH, and band gap energy are the most influential parameters. Overall, alkaline conditions enhance photocatalytic efficiency, and ML proves to be a powerful predictive tool.