A Novel Fisher Discriminant Approach Based on Genetic Algorithm
Kang Jiang, Han Zhao, Zhenhua Yu, XU Lin-shen, Bingyu Sun · 2006
Fisher Linear Discriminant (FLD) is often used in pattern recognition to separate samples from different clusters in multidimensional "feature" space. A novel kernel Fisher discriminant (KFD) method was proposed based on Genetic Algorithm (GA) which can be used to attain the optimal Fisher direction vector. In our approach, the number of parameters that we should find equates to the dimension of training samples instead of the number of training ones. So the computational complexity can be significantly simplified compared with traditional non-GA KFD method. In addition, the selection method for kernel functions was also analyzed and discussed in this paper. Finally, the numerical results verify the effectiveness and efficiency of our approach.