Study Of Test Data Analysis Based On Multivariate Statistics And Complex Network Methods

With the "Big Data" coming, Data Mining has become a hot topic in all walks of life. It can extract valuable information from a large number of disorganized data sets and successfully applied in commerce, finance,healthcare, etc. In recent years, with the development of educational informationization, more and more experts and researchers have focused on the data in education field, and formed a new discipline - "educational data mining". At present, a large number of test scores data stored in the education sector and in the examination institutions are not being fully utilized. Analysis of test data can not only provide advices to education sector and policy makers, evaluate the learning system, but also improve teaching methods, guide and modify students' learning.In this paper, multivariate statistics and complex network method are used to analyze the large-scale test scores data, concluding complex network analysis, factor analysis and multiple regression analysis. First of all, we constructs the spatial structure of science comprehensive examination with the aid of complex network method. It depicts the relationship among physic, chemistry, biology and all examination questions. We finds the community structure in the network of minimum spanning tree and explores the association of the examination ability.Secondly, based on the method of factor analysis in statistics, we analyze the score data of Beijing College Entrance Examination. Combining with classic measures of tests, we find four common factors which reflect the students' ability of comprehensive examination of science. Moreover, we explore a multiple evaluation measurement by the factors score. Finally,we established the multiple regression model to predict the discrimination of the examinations according to the knowledge point and ability point then formed the characteristic property of examinations. It can provides the suggestion to propositional person according to the prediction of the discrimination degree. The data model constructed in this paper is enlightening for quantitative research of knowledge and ability.