Experimental and machine learning analysis on Mn doped Fe2O3@Bi2S3 heterojunctions for visible light photocatalytic degradation of Rhodamine B

In this study, Mn-doped Fe 2 O 3 @Bi 2 S 3 nanocomposite heterojunctions were synthesized, characterized, and evaluated for their photocatalytic performance in the degradation of Rhodamine B (RhB) dye under visible light. A strong heterojunction structure was formed by combining Fe 2 O 3 and Bi 2 S 3 semiconductors through a hydrothermal method, and a series of composites were prepared by incorporating 1–10% Mn dopant into the Bi 2 S 3 phase. Structural (XRD) and optical (UV-Vis DRS) characterization results confirmed the successful formation of the Fe 2 O 3 @Bi 2 S 3 heterojunction and revealed that Mn doping was integrated into the crystal structure, modifying its band configuration. Controlled Mn doping altered the optical properties of the nanocomposite, tuning the band gap between 1.24 and 2.0 eV and thereby optimizing visible-light absorption and charge carrier dynamics. Photocatalytic tests demonstrated that optimal Mn content (particularly 5% Mn) significantly enhanced RhB degradation efficiency. The Fe 2 O 3 @5% Mn–Bi 2 S 3 nanocomposite exhibited the highest photocatalytic performance, achieving approximately 80% RhB removal within 60 min. This improvement was attributed to the efficient charge carrier separation facilitated by the Fe 2 O 3 –Bi 2 S 3 heterojunction and the shallow trap levels introduced by Mn doping, which suppressed recombination. In contrast, excessive Mn doping (>5%) increased carrier recombination via deep trap states, reducing the activity. The novelty of this study lies in the synthesis and performance reporting of Mn-doped Fe 2 O 3 @Bi 2 S 3 heterojunction photocatalysts, which have not been thoroughly investigated in the literature, as well as the integration of machine learning (ML)-based models into the material development process. Various ML algorithms were employed to successfully model RhB degradation data, with CatBoost and XGBoost models providing the most accurate predictions ( R 2 ∼0.99). Furthermore, SHAP analysis provided interpretable insights into the contributions of operational and photocatalyst-related features to the model predictions, distinguishing the influence of reaction conditions from the effects associated with Mn doping ratio and photocatalyst type. Thus, combining experimental findings with ML-driven insights highlights the potential of data-driven approaches for the design of next-generation photocatalysts.