Multi-label learning method based on ML-RBF and laplacian ELM

Multi-label data widely exist in the real world, and the multi-label learning deals with the problem in which samples contain many labels. The main task of the multi-label learning is to train a model which adapts to the multi-label data such that the label of unknown label data can be predicted. Multi-label radial basis function neural network (ML-RBF) is an effective multi-label learning model, which combines K-means clustering and RBF neural network. The laplacian extreme learning machine (Lap-ELM) improved the traditional ELM by considering the structural relationship between low-dimension data and high-dimension data. As a kind of single-hidden layer feed-forward neural network (SLFN), ELM has the characteristics of fast training and good generalization ability compared to RBF. Affinity Propagation (AP) clustering algorithm can automatically determine the number of clusters. In this paper, a novel multi-label learning method named ML-AP-RBF-Lap-ELM is proposed which integrates AP clustering algorithm, ML-RBF and Lap-ELM. In this new model, the ML-RBF is used to map in the input layer. The number of hidden nodes and the center of the RBF function can be automatically determined by the AP clustering algorithm. The weights from the hidden layer to the output layer are solved by Lap-ELM. The simulation results show that ML-AP-RBF-Lap-ELM performs well on the three common data sets, including Natural Scene, Yeast Gene and 20NG (20 New Groups). (C) 2018 Elsevier B.V. All rights reserved.