Incremental learning of feature space and classifier for on-line pattern recognition

Abstract

In the previous work, we have proposed a new approach to pattern recognition tasks, in which not only a classifier but also a feature space is trained incrementally. To implement this idea, Incremental Principal Component Analysis (IPCA) and Evolving Clustering Method (ECM) were effectively combined. However, the original IPCA only gives a way to determine the increase of a new feature based on a threshold value, whose value must be optimized for different datasets. In this paper, to alleviate the dependency on datasets, the accumulation ratio is introduced as its criterion, and an improved algorithm of IPCA is derived. To see if correct feature construction is carried out by this new IPCA algorithm, the classification performance is evaluated over some standard datasets when Evolving Clustering Method (ECM) is adopted as a prototype learning method for Nearest Neighbor classifier. Our simulation results show that the proposed IPCA works well without elaborating sensitive parameter optimization and its recognition accuracy outperforms that of the previous model.