For improving the application performance of Logistic regression models on classification problems, this paper develops a double adaptive elastic net by combining the adaptive Lasso and adaptive Ridge. The double adaptive elastic net has both the oracle property and the adaptive grouping effect, which ensures that it can effectively estimate parameters and accurately select important variables under certain assumed premises and consequently, makes the established Logistic regression model simple and precise. Simulation and case analysis show that the double adaptive elastic net is suitable for medium or high correlation cases with adaptive grouping effect, and its performance of improving Logistic regression is equal to or better than that of the elastic net and other partial improvement methods.