Abstract: Predictive modeling is one of the concepts to find a pattern or a learning model for the next test data. One implementation of this modeling is the decision tree concept. Data used in the simulation is vacant land data. Indicator analysis was conducted to determine patterns or learning models produced from test results using predictor variables towards dependent variable as seen from variable selection as the root and number of variables. Thus, it can be obtained a result that number of variables that used affect the pattern or learning model resulted. Capturing the root to obtain the decision tree does not affect learning model that obtained, so any variable that is used as a root produces the same learning model. The accuracy of variable selection also affects the patterns or learning models resulted. The fewer and inaccuracy in choosing the predictor variables affect the pattern or learning model resulted. Therefore, determination of the used variables must meet the principles of validity.