Authors: Lin Sun, Lanying Wang, Weiping Ding, Yuhua Qian, Jiucheng Xu

Abstract:

For heterogeneous datasets containing numerical  and symbolic feature values, feature selection based on fuzzy  neighborhood multigranulation rough sets (FNMRS) is a very  significant step to preprocess data and improve its classification  performance. This paper presents an FNMRS-based feature  selection approach in neighborhood decision systems. First, some  concepts of fuzzy neighborhood rough sets and neighborhood  multigranulation rough sets are given, and then the FNMRS  model is investigated to construct uncertainty measures. Second,  the optimistic and pessimistic FNMRS models are built by using  fuzzy neighborhood multigranulation lower and upper  approximations from algebra view, and some fuzzy neighborhood  entropy-based uncertainty measures are developed in information  view. Inspired by both algebra and information views based on  the FNMRS model, the fuzzy neighborhood pessimistic  multigranulation entropy is proposed. Third, the Fisher score  model is utilized to delete irrelevant features to decrease the  complexity of high-dimensional datasets, and then a forward  feature selection algorithm is provided to promote the  performance of heterogeneous data classification. Experimental  results on twelve datasets show that the presented model is  effective for selecting important features with higher stability of  classification in neighborhood decision systems.

Keywords： Fuzzy neighborhood rough sets; neighborhood multigranulation rough sets; feature selection; neighborhood entropy; uncertainty measure

Feature selection using fuzzy neighborhood entropy-based uncertainty measures for fuzzy neighborhood multigranulation rough sets.pdf

查看详情

Thu Mar 04 06:30:00 CST 2021