Analysis And Research On The Selection Of Postgraduate Training Indicators Based On Mutual Information

In recent years,with the rapid development of information technology and the advancement of college informatization,colleges and universities have generated a large amount of data in the process of postgraduate training,and these data are becoming increasingly complex and diverse.On the one hand,the abundance of data provides a very valuable data resource for the related research work of postgraduate training,and on the other hand,it also brings new challenges to the related research work.In the analysis and research of postgraduate training in colleges and universities,if the data is directly analyzed,it will not only increase the complexity of the constructed model,affect the performance of the algorithm,but may even cause dimensional disaster.Effectively screening the characteristic indicators of the data generated in the postgraduate training process and removing redundant indicators can greatly reduce the amount of calculation in the process of postgraduate training performance analysis and save computing time.Feature index selection refers to the process of selecting appropriate feature subsets from the original data set to optimize feature evaluation results.The selection of feature indicators has always been a very important link in data mining and big data analysis.In the process of index selection,how to determine the strength of the correlation between the characteristic indicators and how to screen the effective characteristic indicators are two main steps.This paper analyzes the relevant theories and methods of postgraduate performance training at home and abroad,as well as the methods that are widely used to measure the correlation of indicators,and proposes a new indicator screening method.The method is mainly carried out from two directions:improving the correlation measurement standard between feature indicators and optimizing the feature subset screening process.The main work of this paper is summarized as follows:(1)The significance and value of this research,as well as the research status of postgraduate training in colleges and universities at home and abroad,are described in detail.The basic principles and knowledge of mutual information,neighbor propagation clustering algorithm,and neighborhood mutual information required by the project are introduced.(2)Combining the correlation principle of mutual information and the advantages of neighbor propagation clustering algorithm,a new index correlation measurement method,neighbor propagation clustering mutual information APMI,is proposed.In five open source datasets,the multi-dimensional comparison between the clustering mutual information of neighbor propagation and the neighborhood mutual information fully proves that the clustering mutual information of neighbor propagation can be used to measure the correlation between feature indicators.(3)A new index screening method GAPMI is proposed on the basis of the clustering mutual information of neighbor propagation.This method not only considers the correlation between the feature index and the decision index in the data set,but also considers the mutual influence between the feature attributes.In the process of index screening,GAPMI will calculate the comprehensive redundancy and comprehensive information contribution of each candidate feature index,and use the greedy search method to screen.By comparing GAPMI with FCBF,NRS,MNMI and other screening methods,the effectiveness of the screening method proposed in this paper is proved.(4)The verified index correlation strength measurement method APMI and characteristic index screening method GAPMI are applied to the screening of graduate training indicators.In the screening process,the analytic hierarchy process was used to quantify the output results of the multi-index evaluation,and the quantified results were used as decision indicators for screening.Finally,7 feature indicators are screened from 18 input indicators as feature subsets.(5)Finally,the paper gives a detailed summary of the work done in this paper and related research results,points out the shortcomings of the method proposed in this paper according to the problems encountered in the experiment,and looks forward to the future research direction.The method proposed in this paper is based on mutual information,and mutual information is a measurement method based on probability statistics,so as the amount of data increases,this measurement method will become more accurate.Therefore,the method proposed in this paper has certain theoretical significance and practical value for the analysis and research of current postgraduate training.