Attribute Sorting Method Based On Granularity And Its Application In Rough Set

With the rapid development of information technology,the volume and size of data is growing exponentially.Not only is this data growing rapidly,but the dimensionality of the data is also increasing.As people’s lives become more and more informative and efficient,these complex and redundant high-dimensional data are gradually filling up people’s lives.As an effective feature selection technique,attribute simplification has received a lot of attention in recent years.However,some limitations of the existing algorithms are becoming more and more obvious.The large amount of high-dimensional and complex data leads to the problem that the existing attribute simplification algorithms are too time-consuming in the process of data processing and analysis,and the implied information is difficult to be mined,resulting in a bottleneck in the development of traditional attribute simplification methods.In order to overcome the limitations of existing algorithms,this thesis explores the relationship between granularity and attribute reduction process from the perspective of attribute granularity,and improves the traditional simplification algorithm by using the correlation between granularity and attribute discriminative performance;in order to fully exploit the implicit information of the data itself,the granularity-based attribute sorting granularity acceleration optimization strategy is proposed in conjunction with the adaptive characteristics of granular rough set data.In order to fully exploit the data’s own implicit information,we propose a granularity-based granular sphere acceleration optimization strategy,which can significantly improve the time efficiency of the approximate solution while ensuring better classification performance of the algorithm.Specifically,the research contents and results of this thesis include:1.An accelerative solving reduct strategy based on granularity.In the process of deriving reduct by using the forward greedy searching strategy,it is necessary to evaluate all of the candidate attributes,and then one optimal attribute can be selected for each iteration.However,if the number of attributes increases dramatically,then the process of such strategy is very time-consuming.In view of this,following the relationship between the process of forward greedy searching and the variation of granularity,an accelerative solving reduct strategy based on granularity is proposed.The core of such strategy is to remove some attributes,which are corresponding to coarser granulation results based on the size of the granularity,and then the searching space of candidate attributes can be compressed for improving the efficiency of searching reduct.The experimental results show that: on 12 groups of UCI data sets,using three kinds of uncertainty measures,from the perspective of average time consumption,compared with the forward greedy search,the accelerated solution reduction strategy based on granularity can improve the time efficiency of reduction solution by 39.40%,33.49% and 32.07%,respectively,in the case of using approximate quality,conditional entropy and neighborhood discrimination index as measures to construct constraints,At the same time,it can also ensure that the reduction still has good classification performance,which verifies the effectiveness of the proposed algorithm.2.Attribute sorting based quick strategy for searching reduct based on granular ball rough set.An inherent problem with the forward greedy search strategy is that in its neighborhood partitioning process,a radius needs to be artificially set for controlling the size of the neighborhood.To this end,the author introduces the concept of granular spheres,a method that adaptively generates information granules based on the distribution of the data itself.However,in the process of searching reduct,the algorithm still needs to evaluate all candidate attributes one by one to determine whether they can be added to the simplification set.This means that there is still a large search space and there are more redundant attributes in the candidate pool corresponding to a coarser granularity,which makes the search time consumption for the simplification unacceptable when the number of attributes increases dramatically.In view of this,based on the basis of granular sphere rough set simplification solution,we propose an accelerated simplification solution algorithm based on granular sphere rough set attribute sorting based on the relationship between attribute simplification process and granularity,and apply it to the granular sphere rough set model by combining the granularity-based attribute sorting method.The experimental results show that the introduction of the granularity-based attribute ranking method can significantly improve the time efficiency of the algorithm compared with the granular rough set-based simplification search strategy on 16 UCI datasets,and the average time consumption is reduced by 6.5times,thus verifying the effectiveness of the proposed algorithm.On the eight UCI datasets with the addition of noisy data,the proposed algorithm can still guarantee the good classification performance of the resulting simplification while significantly improving the time efficiency and stability of the solved simplification.