PEnSVM: An Improved SVM Ensemble Algorithm Based on Different PCA Thresholds

摘要

In this study, a Support Vector Machine (SVM) ensemble algorithm, named PEnSVM, is presented aiming to improve the classification accuracy of SVM. PEnSVM includes ten base SVM classifiers which are based on different Principal Component Analysis thresholds. Majority volting strategy is adopted to construct the SVM ensemble of the ten base classifiers. Four UCI data sets and a data set from the Uppsala University are used to measure the performance of PEnSVM. When compared with the famous LIBSVM and ensembleSVM, an up-to-date SVM ensemble algorithm, the experimental results show that PEnSVM outperforms those two algorithms in terms of accuracy and sensitivity by and large.