Abstract:

The selection of variables is very important in order to build the best predictive model. The common variable selection methods are mainly the stepwise regression and all possibility regression method. The stepwise regression can not found the best model usually when establishing the predictive model^[1], especially when dealing with an overfitting data. The all possibility regression needs the great calculating quantity and is time-consuming. If the quantity of independent variables is n, the quantities of all possible models are 2ⁿ. So if n>10 the calculation can not be done by an ordinary computer.