Bagging evolutionary feature extraction algorithm for classification

Abstract

Feature extraction is significant for pattern analysis and classification. Those based on genetic algorithms are promising owing to their potential parallelizability and possible applications in large scale and high dimensional data classification. Most recently, Zhao et al. presented a direct evolutionary feature extraction algorithm (DEFE) which can reduce the space complexity and improve the efficiency, thus overcoming the limitations of many genetic algorithm based feature extraction algorithms (EFE). However, DEFE does not consider the outlier problem which could deteriorate the classification performance, especially when the training sample set is small. Moreover, when there are many classes, the null space of within-class scatter matrix (S[sub w]) becomes small, resulting in poor discrimination performance in that space. In this paper, we propose a bagging evolutionary feature extraction algorithm (BEFE) incorporating bagging into a revised DEFE algorithm to improve the DEFE's performance in cases of small training sets and large number of classes. The proposed algorithm has been applied to face recognition and testified using the Yale and ORL face databases.