Formulating Particle Swarm Optimization based Generalized Kernel Function for Kernel-Linear Discriminant Analysis

Selection of kernel function for solving Kernel-Linear Discriminant Analysis (K-LDA) remains unsolved problem. In this commuication, we propose the method to formulate the Generalized Kernel Function (GKF) for K-LDA.The parameters of the GKF are tuned using the Particle Swarm Optimization (PSO) to maximize the discrimination in the higher dimensional space. Experiments are performed on the petal shaped synthetic toy cluster using the proposed GKF and are compared with the results obtained using the standard kernel functions. The experimental results reveals the importance of using the proposed technique.