Research On Approximate Attribute Reduction Based On Neighborhood Rough Set

Attribute reduction is one of the important research contents in data mining,aiming to remove redundant conditional attributes in the original dataset without reducing the data classification ability,thereby obtaining a more concise description of the original dataset.Rough set theory,as an important mathematical method for describing imprecise knowledge,performs better on symbolic data than on numerical data.To better handle numerical data,neighborhood rough set model and other extended models have been proposed.However,the reduction conditions of attribute reduction algorithms in both classical rough set models and neighborhood rough set models are too strict,and the research on neighborhood rough set reduction models generally focuses on numerical datasets.This paper conducts approximate attribute reduction research on mixed datasets based on neighborhood rough set.The main work is as follows:(1)In response to the inefficiency of classical attribute reduction methods in dealing with mixed datasets,Fast Attribute Reduction based on Dual Optimization for Neighborhood Rough Set(FARDONRS)method is proposed.The new method first preprocesses the dataset by bucketing and then partitions the attribute set to improve algorithm efficiency.Experimental results show that the reduction effect of FARDONRS on mixed datasets is better.(2)Aiming at the subjectivity of neighborhood radius determination in neighborhood rough set,an adaptive neighborhood information system is constructed using the concept of natural neighborhood.Based on this,an attribute separability measure function is proposed to evaluate the ability of attributes to distinguish objects,which synthetically assesses the value of information carried by condition attribute and decision attribute,then,Approximate Attribute Reduction based on Natural Neighborhood System(AARNa NS)is proposed.Experiments on UCI standard datasets show that AARNa NS is more effective in detecting mixed datasets.