A Fast Neural Network Multi-label Classification Algorithm Based on Approximate Extreme Points
Xupeng Wang, Zhongwen Guo, Xi Wang, Mengdi Min, Yangfan Zhao · 2019
In order to solve the problem that the training time of existing neural network multi-label classification algorithms is too long, we propose a new BP neural network multi-label classification (BP-AEPML) algorithm based on approximate extreme points. Firstly, the original multi-label data set is transformed into multiple separate binary sub data sets using binary relevance transformation strategy. And then the representative data set of each binary sub data set is extracted using approximate extreme points. Finally, the representative data set is trained with BP neural network. To validate and evaluate the classification results and training time of the proposed algorithm, three public data sets are used to compared with traditional BP neural network multi-label classification algorithm, basic SVM multi-label classification algorithm and ML-BVM algorithm. The experiment results shows that the classification results of BPAEPML is similar to that of the traditional BP neural network multi-label classification algorithm and a little better than that of basic SVM multi-label classification algorithm and ML-BVM algorithm. At the same time, the training time is greatly reduced.