An Efficient Extraction-based Bagging Ensemble for High-dimensional data classification

In high-dimensional data classification, the method employed should be both powerful and robust against the SSS (small sample size) problem. LDA is a classical, efficient, and powerful feature extraction method that can be applied to effectively reduce the feature space dimension and thus ease the adverse effect of the SSS problem. However, LDA itself suffers from the SSS problem due to the nature of its separability measure. In this study, a modified version of LDA called ARLDA is proposed to efficiently counter the SSS problem of LDA. To increase performance, ARLDA is embedded in a Bagging framework to form a multi-classifier ensemble called EEBBE. The performance of EEBBE is evaluated by experiments based on a hyperspectral image and three UCI data sets. The results showed that EEBBE is a very promising classification method.