A Futuristic\nApproach to Subsurface-Constructed Wetland\nDesign for the South-East Asian Region Using Machine Learning

This study investigates the optimized design of horizontal\nflow\nconstructed wetlands (HFCWs) to enhance pollutant removal efficiency\nwhile minimizing surface area requirements, particularly in the Southeast\nAsian region. By refining the first-order removal rate coefficient\n(<i>k</i>) for organics and nutrients, the research aims\nto meet specific performance benchmarks across three scenarios, ensuring\ncompliance with discharge or reuse standards. Utilizing a data set\ncomprising 1680 entries, five machine learning modelsmultiple\nlinear regression (MLR), eXtreme Gradient Boosting (XGBoost), random\nforest (RF), artificial neural network (ANN), and support vector regression\n(SVR)were employed to predict <i>k</i> values. Pearson’s\ncorrelation, heat maps, and ANOVA analysis identified the most influential\nparameters affecting <i>k</i>-value predictions. The <i>k</i> values ranged from 0.01 to 0.52 per day using the <i>P</i>–<i>k</i>–<i>C</i>* method,\nessential for effective pollutant removal. The SVR model demonstrated\nthe highest predictive accuracy, with <i>R</i><sup>2</sup> values of 0.91 for <i>k</i><sub>BOD</sub>, 0.90 for <i>k</i><sub>TN</sub>, 0.82 for <i>k</i><sub>TKN</sub>, and 0.76 for <i>k</i><sub>TP</sub>. This optimization\nreduced standard deviations significantly, from 136.90% to 2.28%.\nConsequently, the required wetland area was reduced by up to 68% for\nbiochemical oxygen demand (BOD), 60% for TN (total nitrogen), and\n67% for TP (total phosphorus) in larger systems, supporting the tailored\ndesign of HFCWs to meet targeted discharge standards.