Research And Application Of Attribute Reduction Algorithm Based On Neighborhood Rough Set

Rough set theory is a kind of data information processing tool,which can help us to efficiently mine and obtain the information we need from the mass data,and has been widely used in many fields.However,the classical rough set theory model is only suitable for processing discrete data,and when processing continuous data with it,the data needs to be discretized first,which may easily lead to the loss of some data information.For this neighborhood rough set model by introducing the graining neighborhood and the concept of metric space,turn the equivalence relation of rough set theory to information on bounded domain space particles covering relations,can be directly with continuous data,avoid the original information loss of the problems of rough set to deal with continuous attributes,expand the scope of the model.At the same time,because the model of neighborhood rough set introduces the computation of neighborhood granulation,the efficiency of the whole algorithm decreases.On the basis of the existing neighborhood rough set attribute reduction model,this paper improves the problem and verifies it by experiments.In addition,this paper applies the improved algorithm to the improved C4.5 decision tree classifier model for in-depth research.The main work of this paper is as follows:(1)The existing neighborhood rough set attribute reduction algorithm is analyzed.In view of the existing algorithm,when attribute importance is judged by the dependency function to carry out attribute reduction,there are repeated redundant calculations,resulting in high algorithm complexity and large computation amount.This paper redefines the method of attribute importance and reduces the complexity of the algorithm.At the same time,in order to reduce the impact of the correlation between attributes on the final result,the relevant knowledge of the correlation coefficient is introduced to further screen the attributes.Finally,an attribute reduction algorithm for binary classification problem is proposed.By comparing with other algorithms,it is proved that this algorithm can reduce the complexity of attribute reduction and improve the operation efficiency.(2)In reality,there are many multi-classification problems besides dichotomy.In order to overcome the limitations of the above algorithm,this paper proposes a weighted function of attribute importance for multi-classification based on the Relief algorithm,and applies it to the attribute reduction model of neighborhood rough sets.Finally,a fast attribute reduction algorithm for neighborhood rough sets based on Relief algorithm is proposed.Experimental results show that the algorithm is effective and feasible.(3)The classification algorithm of C4.5 decision tree was analyzed and the existing problems were improved accordingly.In this paper,the improved algorithm of attribute importance as C4.5 algorithm split node selection criteria,and at the same time,based on the boundary point theorem Fayyad,optimum continuous attribute threshold segmentation method,reduce the threshold when the choice of data sets,the number of times past,build give a kind of applies to both discrete and continuous data classifier model,and through the experiment proves that the model for improving decision tree classification accuracy and efficiency of the decision tree generation is effective.