The Boosted/Bagged Subclass Method

The subclass method is one of pattern classification methods proposed by Kudo et al. (1989), which is based on the approximation of each class region by a set of axis-parallel hyper-rectangles. This study improved it using an adaptive resampling technique known as boost'ing. Boosting is a well known ensemble learning method as an effective tool for improving the classification performance. Regarding the subclass method as a method for controlling the performance of resultant classifier, we could investigate 1) how the performance of a base classifier effects the classification results by boosting, and 2) how much boosting can improve the results compared with the original subclass method. Moreover, we also investigated the result using bagging which is an another popular ensemble technique