Prediction of hydraulics performance in drain envelopes using Kmeans based multivariate adaptive regression spline

Abstract

This study suggests a new modeling strategy, hybridized multivariate adaptive regression spline (MARS) and Kmeans clustering, for estimating coefficients of hydraulic conductivity using various input combinations of the useful variables, hydraulic head H (cm), geotextile filters size O90 (nm), time T (min) and discharge of drain Q (cm3/s). The results of the newly developed method (MARS-Kmeans) were compared with the single MARS, M5 model tree (M5 Tree) and group method of data handling (GMDH) with respect to four assessing statistics of root mean square errors (RMSE), mean absolute errors (MSE), Nash Sutcliffe efficiency (NSE), and coefficient and determination (R2) together with Wilcoxon rank-sum test and visual evaluation via scatterplots, boxplots, and Taylor diagram. The results indicated the superiority of the MARS-Kmeans method over the M5 Tree, MARS, and GMDH in estimating envelope hydraulic conductivity and soil-envelope hydraulic conductivity. The accuracy of the M5 Tree, MARS and, GMDH methods were improved using MARS-Kmeans in estimating Kse by 45%, 57%, and 77% and estimating Ke by 31%, 38%, and 45% for RMSE.