Multiple Regression Models up to First-order Interaction on Hydrochemistry Properties

This study illustrated the procedure in selecting the best model in estimating the Electrical Conductivity (EC) levels based on the hydrochemistry properties and nature effecting factors using multiple regressions. The six independent variables and two dummy variables considered in this data set. The Multiple Regression (MR) models were involved up to first-order interaction and there were 57 possible models considered. This study is the extension of prior research which had generated 63 possible models, by using the same technique but no interaction involved between the independent variables. In this study, the process of getting the best model from the total of 120 possible models had been illustrated. The backward elimination of variables with the highest p-value was employed to get the selected model. The best model includes the combination of single and first order interaction (Li, Mg, Na-SO4, Na-Li, Na-Mg and SO4-Mg). The best model obtains then being verified by the Mean Absolute Percentage Error (MAPE) calculation to measure the models’ relative overall fit.