Determine the Parameter of Kernel Discriminant Analysis in Accordance with Fisher Criterion

• Published in: 2007 International Conference on Machine Learning and Cybernetics Vol. 5; pp. 2931 - 2935
• Main Authors: Xu, Yong, Li, Wei-Jie
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.08.2007
• Subjects: Computer science; Cybernetics; Design methodology; Feature extraction; Fisher criterion; Iterative methods; Kernel; Kernel discriminant analysis (KDA); Kernel function; Machine learning; Machine learning algorithms; Parameter selection; Pattern analysis; Performance analysis
• ISBN: 1424409721; 9781424409723
• ISSN: 2160-133X
• DOI: 10.1109/ICMLC.2007.4370649

Feature extraction performance of kernel discriminant analysis (KDA) is influenced by the value of the parameter of the kernel function. Usually one is hard to effectively exert the performance of FDA for it is not easy to determine the optimal value for the kernel parameter. Though some approaches have been proposed to automatically determine the parameter of FDA, it seems that none of these approaches takes the nature of FDA into account in selecting the value for the kernel parameter. In this paper, we develop a novel parameter selection approach that is subject to the essence of Fisher discriminant analysis. This approach is theoretically able to achieve the kernel parameter that is associated with a feature space with satisfactory linear separability. The approach can be carried out using an iterative computation procedure. Experimental results show that the developed approach does result in much higher classification accuracy than naive KDA.