A Novel Subspace Classification Method for Large Datasets Based on Kernel-Based Fisher Discriminant Analysis

With the rapid development of E-business and database, classification for high dimensions in large scale datasets becomes an important task of business intelligence. Recently, kernel-based methods have attracted more and more attention and have shown excellent performance in pattern recognition, machine learning and image classification etc. The common weakness of the kernel-based learning algorithms is that they cannot deal with a large dataset. In this paper, a novel classification method for large datasets named Sub-KFDA (Subspace classification based on Kernel Fisher Discriminant Analysis) is presented. A subspace mining approach based on frequent patterns and kernel-based fisher discriminant analysis is designed to decompose the initial large dataset classification problem into many small dataset classification problems. Experiment results on UCI datasets demonstrate that the proposed method has advantages in accuracy in comparison to other classification approaches.