Hierarchical Feature Selection via Large Margin Nearest Neighbor

Being in the era of big data, the complexity of data is getting higher and higher and the well organized data has been the important driving force for machine learning and data mining tasks, which means that raw data need to be processed to satisfy the tasks from different fields. Different classes in hierarchical structure gives rise to the hierarchical classification problem, which varies from the flat classification such as binary or multi-class classification problems in general, and now is naturally ubiquitous due to the rapid growth of data in the real world. Feature selection for hierarchical classification is confronted with the problem that traditional linear regression method cannot classify data well, especially in the context of classes with close relationships. In this paper, we introduce a novel feature selection method based on large margin nearest neighbor strategy (HFSLM), and apply it to the hierarchical classification problems. The key idea of this method is to project the data into a low-dimensional feature space and separate samples from different classes. Experimental results on several hierarchical class datasets show that our method is better than the representative HiFSRR method.