Stream Data Classification Using Improved Fisher Discriminate Analysis

Abstract

A modified Fisher discriminate analysis method for classifying stream data is presented. To satisfy the realtime demand in classifying stream data, this method defines a new criterion for Fisher discriminate analysis. Since the new criterion requires less computation and memory space, it is much faster and more suitable for online processing in stream data environment. It can overcome the problem of singular within-class scatter matrix in traditional FDA. Our algorithm speeds up the mining process while maintaining the high classification accuracy and capturing the up-todate trends in the stream. Experiments on real and synthetic data sets show that our algorithm can improve the classification accuracy and speed for stream data classification.