An automatic fuzzy neighbourhood cutting algorithm for attribute reduction

Abstract Neighbourhood information represents a local component of granular information and is employed to describe the distribution of neighbourhoods surrounding a given sample. Automatically choosing the proper size to determine the granularity of neighborhoods is a challenging problem. To address these issues, we introduce a cutting mechanism from fuzzy set theory to generate fuzzy neighbourhoods with automatic parameters, these parameters are the confidence level determining which samples are selected into the fuzzy neighborhood of the sample. Misclassified samples are identified by their diverse category labels within the fuzzy neighbourhood, and their distributions can reflect the performance of a classification task. Additionally, we propose a reduction algorithm (CFE) to choose a pertinent subset of attributes from the dataset. Experimental findings on actual datasets reveal the algorithm’s superiority over baseline approaches and its efficiency in eliminating misclassified neighbours.