Research On Neighborhood Rough Set Attribute Reduction Algorithm Based On Space Partition

With the highly developed information technology,the quantity and dimension of data have increased dramatically.This inevitably creates redundant data,which results in reduced information density of data and difficulties in storing and using data.As a theory of extracting knowledge and information from data,the main idea of rough set theory is to granulate data through equivalence relation,and use the equivalence class obtained by granulation to approximately describe uncertain concepts.Attribute reduction is a popular application derived from rough set theory,which aims to remove redundant attributes and retain important attributes.Since classical rough set granulates data through equivalence relation,it can only process discrete data,which limits the development and application of this theory.Neighborhood rough set model introduces neighborhood and granulates data according to the relationship between data and other data within its neighborhood,thus overcoming the drawback of classical rough set.However,the neighborhood rough set model also has some problems such as low efficiency and being affected by distance metric.Aiming at the mentioned problems,this thesis conducts the following research:1.Since the neighborhood of neighborhood rough set model is based on the distance metric,the classification effectiveness and computational efficiency of this model are both reduced.To solve this problem,this thesis proposes a new neighborhood rough set model based on space partition.This model avoids using distance metric by partitioning the data to obtain a spatial hierarchy and defining a neighborhood based on the spatial hierarchy.Based on this model,this thesis designs a forward attribute reduction algorithm.The experimental results on multiple datasets show that,compared with other cutting-edge neighborhood rough set models,the proposed algorithm has advantages in both efficiency and effectiveness.2.Due to some problems in the stability of the space partition process of the attribute reduction algorithm established in this thesis,the performance of the algorithm is degraded.In this thesis,an adaptive multiple space partition improved method is proposed by introducing a voting mechanism.This improved method reduces the randomness of space partition,and it can adaptively set up the voting scale.The experimental results on the same datasets show that,the performance of the improved algorithm increase due to the improved stability.And compared with the original algorithm,the efficiency of the improved method does not significantly reduce.