Recursive dimensionality reduction using Fisher’s linear discriminant

摘要

Dimensionality reduction is an important part of the pattern recognition process. It would be very useful to have a recursive form for dimensionality reduction that is suitable for implementation on massive data sets and real-time automatic pattern recognition systems. It would also be beneficial to have a version where the dimensionality reduction can be updated based on new partially identified data that are obtained in real systems. Versions of Fisher’s Linear Discriminant for dimensionality reduction that address these problems are derived in this article.