The Technology of Selective Multiple Classifiers Ensemble Based on Kernel Clustering
Xianyi Cheng, Guo Hong-ling · 2008
Because of the high request to classifiers performance of people and the implementation complexity of multiple classifiers ensemble approach, this paper proposes a new method of selective multiple classifiers according to the distribution characteristic of classifiers, the classifying performance as well as the existence diversity. The algorithm uses the Kernel-based clustering method to estimate the performance of each classifier in the whole feature space .And we choose the classifiers which has diversity to form the last ensemble classifier set according to that each classifier has diversity. We carry on contrasting experiments to compare the method which we propose with the bagging method and the best method in the ELENA data set. From the result of theoretic analysis and experiment, we could see that the classifiers ensemble method is efficient in pattern recognition field.