Research On Online Learning Algorithms Based On Kernel Discriminant Analysis For Data Streams

Data stream analysis aims at extracting discriminative information for classification from continuously incoming samples.It is extremely challenging to detect novel data while incrementally updating the model in an efficient and stable fashion,especially for high-dimensional and/or large-scale data streams.It is therefore necessary to incrementally add data from the data stream to the training model instead of retraining all the data by batch learning.This paper proposes an algorithm for novelty detection,incremental learning for unlabeled chunk data streams.Based on the basic principle of linear discriminant analysis,a fast factorization-free kernel discriminative analysis(FKDA-X)is put forward through solving a linear system in kernel space.FKDA-X produces a reproducing kernel Hilbert space in which unlabeled chunk data can be detected and classified by multiple known-classes in a single decision model with deterministic classification boundaries.This method solves the huge time complexity and space complexity problem caused by the large amount of matrix calculation in other existing methods.Moreover,based on FKDA-X,an updated FKDA-CX is proposed.By using micro-cluster centers of original data as input,FKDA-CX has excellent performance in novelty detection.Both theoretical analysis and experimental can show that the proposed algorithm makes it possible to learn unmarked block data streams,and it has big advantages on accuracy and time complexity compared with the state-of-the-art methods.