Integrated Analysis of a Tower Filtration–Constructed Wetland System Using Waste Brick Fillers and Random Forest Regression for Rural Wastewater Treatment

Rural domestic wastewater treatment often suffers from high costs and poor adaptability to fluctuating influent conditions. To overcome these challenges, this study developed a combined system integrating a tower biological filter and a vertical subsurface flow constructed wetland, using recycled waste bricks as the main filler. The objectives were to evaluate pollutant removal efficiency and explore the influencing factors through both experimental and data-driven analyses. Adsorption kinetics and isotherm tests were conducted to assess the pollutant adsorption characteristics of waste bricks. Meanwhile, a Random Forest regression model was introduced to identify key variables affecting removal efficiency and to reveal nonlinear relationships between operational parameters. The model incorporated factors such as influent concentration, pollutant type, filler dosage, particle size, and hydraulic loading, with results visualized through feature importance, partial dependence, and heatmap analyses. Experimental findings showed strong adsorption of NH₃–N and TP, with maximum capacities of 0.0669 mg/g and 0.0397 mg/g. Under an optimal hydraulic loading of 0.3 m3/(m2·d), the system achieved average removal efficiencies of 90.8% (COD), 98% (NH3–N), 47% (TN), and 90% (TP). The integrated process demonstrates high stability, low cost, and adaptability, while the machine learning approach provides valuable insights for mechanism interpretation and performance optimization.