Machine learning powered prediction of photodegradation of 2,4-dichlorophenoxyacetic acid using gold-doped bismuth ferrite

This study investigates the potential of various tree-based machine learning models to predict the photodegradation of 2,4-dichlorophenoxyacetic acid (2,4-D) from wastewater using different gold-doped bismuth ferrite (Au-BFO) nanocomposites. A detailed dataset comprising 1,050 data points was compiled under a range of experimental conditions. The fourteen input variables included: time (min), catalyst loading (g/L), solution pH, initial 2,4-D concentration (mg/L), anion type, light intensity (W), photocatalyst type, bandgap energy (eV), surface area (m²/g), pore volume (cm³/g), and the contents of bismuth (%), iron (%), oxygen (%), and gold (%) were used for model building. The photodegradation efficiency of 2,4-D was the output variable. Performance evaluation indicated that the XGBoost model was the most effective, achieving the highest test R 2 (1.0) and the lowest values for MAE (0.99) and RMSE (1.88). The results revealed that the photocatalytic experimental conditions were of greater significance compared to the elemental composition and physicochemical properties of the photocatalysts. The optimal photocatalytic conditions identified were the catalyst loading of 1.5 g/L, light intensity of 105 W, initial 2,4-D concentration of 5 mg/L, and a solution pH of 7. The study concludes with a discussion of its practical implications and limitations. • A dataset of 1050 data-points with 14 input features were for ML modelling. • 15 machine learning models were applied to predict the photodegradation of 2,4-D. • ML findings suggest that the XGBoost model outperformed other models. • SHAP analysis proposed that the time, C i , solution pH, and anions were the most influential inputs. • The applications and drawbacks of the current study were stated in detail.