Classification and regression tree technique (CART): The alternative to the multiple regression and associated techniques or methods for determining subsets of the explanatory variables most significant for prediction of the response variable. Rather than ?tting the model to the sample data, a tree structure is obtained by dividing the sample recursively into the various of sets, each division being chosen so as to maximize some measure of difference in the response variable in the resulting two sets. The resulting structure often gives us the easier interpretation than a regression equation, as those variables most significant for the prediction can be quickly identi?ed. In addition this approach does not need distributional assumptions and is also more resistant to the effects of the outliers. At each stage the sample is divided on the basis of a variable, xi, according to answers to such questions as 'Is xi c' (univariate split), is ' Paixi c' (which is linear function split) and 'does xi A' (if xi is the categorical variable). A design of the application of this method or technique is shown in the figure 35.