Feature Selection Based on Kernelized Fuzzy Rough Set for Hierarchical Classification

Feature selection is an effective approach to mitigate the large scale of data in classification learning. However, there typically exists a hierarchical structure in data. Within this structure, features that offer distinct information for different categories are frequently overlooked. In this paper, we took this difference into account. Firstly, we utilize the hierarchical structure of data to divide classification task into various distinct subtasks. Secondly, the feature distinctions in each category are considered and the hierarchical category dependency is defined. Finally, a new feature selection framework is designed based on the hierarchical category dependency. Compared with the other three feature selection algorithms on four datasets, the experimental results prove the superiority of the algorithm proposed in this paper.