A comparative analysis of MLR, MARS, and GMDH techniques for predicting the recharge rate of a sand-mixed storm runoff filtration system

ABSTRACT The current study focuses on assessing effectiveness of filtering systems for different storm runoff influent silt concentrations. In this study, filter media had three layers: at top coarse sand (CS), layer after gravel (G) in middle, and boulders layer at bottom (P) of filtration system. The CS sizes used were namely 0.150–0.300 mm (T-1), 0.300–0.425 mm (T-2), and 0.425–0.600 mm (T-3). The recharging rate and filtering efficiency of CS were observed under seven types of influent sediment concentrations ranging from 250 to 3,000 ppm of storm runoff. Additionally, multiple linear regression (MLR) analysis, multivariate adaptive regression spline (MARS), and group method of data handling (GMDH) soft computing techniques were also employed to predict the recharge rate of filtration system. The MARS model outperformed MLR and GMDH models. The MARS model achieved R2, RMSE, and NSE values for training 0.976, 0.006, and 0.976, and for validation, 0.958, 0.008, and 0.957, respectively.