Selective SVM Ensemble Based on Improved Spectral Clustering

Selective svm ensemble algorithm based on improved spectral clustering is presented to improve the forecasting accuracy and generalization performance of svm effectively. Many svms are produced by bootstrap method and all svms is clustered using spectral clustering based on mutual information, Generalization error function is constructed based on negative correlation learning and svms of generalization error minimum is ensembled by weighted, which ensures accuracy of individual svms and improves the diversity of the individual svms. Experiments results show the algorithm improves the generalization ability of svm and ensemble efficiency, it is an effect ensemble method.