The accurate prediction of environmental Se (selenium) adsorption levels is critical for sustainable development and management perception. The concept of sorting massive quantities of data to find important information using machine learning (ML) has recently been applied to environmental remediation, particularly the science-based design of a 'green' carbonaceous and an effective functional material (e.g., biochar and Fe modified biochar) with high Selenium (Se) removal capacity. The present study focuses on presenting ML models that utilize the random-forest (RF) support vector regression (SVR) and SHAP (SHapley Additive exPlanations) models to forecast the adsorption of Se by modified biochar. The RF, SVR and SHAP models, which were constructed using basic surface properties of Fe-modified biochar and environmental conditions showed accuracy and predictive performance for Se (removal capacity in the test group with R² of 0.98, 0.98 and 0.95 and RMSE of 0.35, 0.14 and 0.23 mg-kg⁻¹, respectively). The SVR model was highly effective for predicting Se adsorption, indicating potentially higher accuracy than the RF and SHAP models. This may be due to the small size of our data. According to the feature analysis and partial dependence plot analysis of all three models, the most significant component regulating Se adsorption was oxygen (%) followed by carbon (%), temperature, pH and Fe in all three models. The relative importance of variables may offer guidance for researchers to develop improved Se treatment of actual water and wastewater. Moreover, the ML models developed in this study took into account the surface functionalities of biochar and Fe-modified biochar to provide a more accurate prediction of Se removal, and offer a comprehensive guideline for the long-term development of biochar adsorbents for Se removal. Graphical Abstract