Reliable protocol using gradient boosting decision tree with limited experimental data to modify membrane surface for enhanced resilience and nitrogen removal in biofilm system

This study highlights the importance of advanced membrane modification optimization methods for Membrane biofilm reactors (MBfR). Using Response surface methodology (RSM) and Gradient Boosting Decision Tree (GBDT), a relationship between solution concentration, reaction time, and temperature versus membrane surface characteristics was established. The GBDT model accurately predicted surface roughness during training (R² = 0.994) and testing (R²_pred = 0.993), outperforming RSM (R² = 0.981, R²_pred = 0.763, RMSE = 8.302) with a low RMSE of 2.077. The optimal conditions (1 mg·L^-1 concentration, 5 h reaction time, 45 ℃) were used to modify membranes, which enhanced bio-attachment greatly. The optimized membrane achieved a nitrogen removal efficiency of 98.4 % in MBfR, performs better than the control one of 35.7 %. These results demonstrate that GBDT shows promising potential in optimizing membrane modification parameters, providing an effective strategy for enhancing MBfR performance through modelling and modification methods.