关键词: 机器学习, 支持向量机, 随机投影, SMO算法, 降维

Abstract: Classification accuracy of large-scale SVM will decline after random-projection-based feature dimension reduction. To remove this shortcoming, Drp-LSVM against large-scale classification problems was proposed in this paper with an introduction of the dual recovery theory. For drp-LSVM, its relevant geometric properties were analyzed and demonstrated here, in aim of proving that Drp-LSVM has maintained the similar geometric advantages of rp-LSVM, and its divided hyperplane is more closer to the primitive classifier trained by complete data. As to the fast solution to drp-LSVM, traditional SMO algorithm was given and a drp-LSVM classifier based on improved SMO algorithm was completed. Finally, it was proved that drp-LSVM has inherited the advantages of rp-LSVM, reduced classification error, improved training accuracy, and all its performance indexes are more close to the classifier trained by primitive data; The classifier designed based on the improved SMO algorithm can reduce memory consumption, and achieve better training accuracy.

Key words: machine learning, Support Vector Machine(SVM), Random Projection, Sequential Minimal Optimization(SMO), dimensionality reduction