The Study On Academic Records Prediction And Students’ Friendship Network Detection Based On Consumption Data Of Campus Card

With the further development of university informatization, the campus card system has pervade most aspects of students’ daily lives. Massive data generated from the students’ activities has been specifically recorded by the campus card system on this background. The data makes it possible for us to analyze the students’ behavior, but how to exploit the huge amount of data to provide a more scientific management of university for the administration office is a challenging problem for us.So far, the researches based on campus card consumption data have been mainly focus on using the data mining techniques or statistical analysis methods to find out the consumption behavior of the students. Most students spend their time enjoying their college life and working hard for their academic achievements. Thus the most important behaviors of university students are communication behavior and learning behavior.However, the traditional method using the questionnaires to study the students’ behavior has defaults that it is usually lack of objective evidence. So whether it is possible to study the communication behavior and learning behavior is what we really cares.As the respective results of communication behavior and learning behavior, the academic records and the friendships are important reflections of them. Therefore, it is a good way to study communication behavior and learning behavior by study the academic records and the friendships first.This thesis has studied the academic records and the friendships using the complex network method and the statistical learning method based on consumption data.The main works of the thesis are as follows:(1) The correlation between the breakfast frequency and the academic records of the students of different majors and grades has been studied. Also the correlation between the burstiness parameter of the students’consumption and the academic records has been analyzed. Then we utilize the KNN(K Nearest Neighbors) classification algorithm based on the two features above to make a prediction for the students’ academic records level. The result is rather impressive.(2) A bipartite network is constructed based on the consumption data. Then the multiple tests method has been applied to statistically validate the co-occurrence relationship of the students. Thus the students’ friendship network is attained and the basic properties of the friendship network are analyzed.