One of the most important issues in classification is to improve the accuracy of the prediction model. Therefore, many researches have focused on this issue recently. Ensemble method that usually referred as 'perturb and combine' strategy has gained g...
One of the most important issues in classification is to improve the accuracy of the prediction model. Therefore, many researches have focused on this issue recently. Ensemble method that usually referred as 'perturb and combine' strategy has gained great popularity. In this thesis, a new ensemble method called 'Logistic Bagging' is proposed, then compared with other methods such as Bagging, Double Bagging, and Logistic regression using real data set and simulations. It was found that the Logistic Bagging method outperformed Bagging and Logistic regression. It also showed better accuracy than Double Bagging for non-normal data. However, for normal data, Logistic Bagging and Double Bagging method didn't show significant difference. Finally, the bias-variance decomposition revealed that the new method reduced the variance more compared to ensemble methods such as Bagging and Double Bagging. On the other hand, it had similar variance but smaller bias when compared to the Logistic regression.