Incremental Feature Extraction Based on Gaussian Maximum Likelihood

Abstract

Linear feature extraction is used to lower data dimensionality using linear projection while minimizing information loss. Most previous algorithms have used the batch mode that process all the available data at once. However, incremental algorithms are also important to rapidly process real-time streaming data. Decision boundary feature extraction (DBFE) is a batch feature extraction method that uses the decision boundaries which a classifier defines. In this paper, an incremental gradient descent DBFE (IGDDBFE) is proposed for Gaussian maximum likelihood classifier. The proposed method showed better performance than other existing incremental feature extraction methods when applied to real-world UCI databases.