Multi-label learning by LLE dimension reduction and Bayesian classification
Min Wu · Systems engineering and electronics · 2009
Samples of multi-label data may belong to more than one class,so its classification problem is much more complicated than single-label data.A novel multi-label learning algorithm is proposed.The feature attributes of multi-label data often have high dimensions,so an LLE algorithm is applied to decrease the dimension in order to extract a group of low dimensional feature attribute sets which could completely describe data.Then multi-label samples are partitioned in terms of their belonging classes,and the classification characteristics of each group are learned by using Bayesian classification model.After that,the final class-label set of multi-label samples is obtained according to the decision class-label of each classification model.The algorithm is applied to the multi-label classification learning of both nature scene image and gene data respectively.Experimental results show that the proposed algorithm can acquire good classification effects on different multi-label datasets and has better performance compared with the others.