The application of Residual Augmented Least Squares method to predict the consistency properties of special clayey soils

Abstract

In this paper, we demonstrate the implementation of a new regression method, Residual Augmented Least Squares (RALS), to predict the consistency properties of special clayey soils. The RALS is a statistical method that is used to model a linear relationship in the case of the non-normal distribution of residuals in linear regression. The method has its roots in the field of econometrics, and in this paper, we demonstrate that the RALS method can be successfully applied for efficiently and accurately modeling the relation between the plasticity index (PI) and the liquid limit (w(L)) of clayey soils when the residual normality assumption of linear regression was not met. In this study, 400 soil investigation reports were used to form a new database that will be used to define the characteristic properties of special soils of Istanbul. The dataset formed in this study contained 2890 liquid limit test and plastic limit test results that were obtained from the field investigation reports. The dataset consisted of two subsets as high plastic clayey soils (CH) data, low plastic clayey soils (CL) data, and a combined dataset (of CH and CL data). A linear regression analysis has been made in the first stage to model the relationship between PI and w(L). But, it should be noted that the different percentages of the evaluated data have been removed from its dataset during the analysis as they were found as outliers based on box-whisker plots. The residuals of linear regression(s) did not meet the normal distribution assumption. Thus, in the next stage, RALS-based regression analyses have been conducted to model the relationship more reliably. The results of both linear and RALS-based analyses showed that RALS-based regression analysis provides more accurate results when compared with linear regression, while also being more reliable in regression analysis with the non-normal distribution of residuals.