No abstract

No abstract

No abstract

All covariates were centred and scaled before running the feature selection. The univariate feature selection (UFS) fits a model for many iterations after filtering of the covariates. Covariate filtering selects covariates by determining the correlation of individual covariates to the soil property of interest. The robust feature selection (RFS) takes all the covariates and progressively eliminates each until the error rate reaches an optimal level. No tuning parameters were optimized for either technique.The least absolute shrinkage and selection operator (LASSO) is a generalized linear model which minimizes covariate coefficients based on the absolute error of the residuals (𝐿𝐿 1 regularisation) through coordinate descent (Friedman et al., 2010). This process shrinks covariate coefficients which are correlated to one another. The degree of shrinkage is controlled by the λ value which was optimised and the covariates which did not have an absolute value of zero, were selected. A LASSO feature selection was