Academic Performance Prediction And Behavior Analysis Based On Campus Big Data

The popularization of digital campus not only brings great convenience to the life of teachers and students,but also provides abundant data for the research of educational data mining(EDM).EDM adopts data mining,machine learning and other technologies to analyze various data records of students in school,so as to solve the problems related to education and provide good services for teachers and students in Colleges and universities.There are many kinds of data related to students,so the paper will make full use of campus big data to analyze students' behavior in school,and realize the academic performance prediction model,and provide valuable reference information for the assistants in order to make correct decisions.Academic performance prediction is one of the most popular research fields in EDM.At present,many scholars have carried out relevant research.However,most of the research methods are single,mainly extracting behavior features through feature engineering,and then using machine learning method to model data.Traditional machine learning methods are lack of innovation,and ignore the hidden relationship between students.In view of the above problems,this paper proposes Dis_GAT(Distance Graph Attention Network)network based on graph neural network to solve the problem of academic performance prediction,and compares it with traditional machine learning methods to prove the effectiveness of Dis_GAT network.In addition to predicting academic performance with multiple data sources,this paper analyses students' living conditions in school from the aspects of consumption behavior and offline friendship behavior,so that counselors can know students' living conditions in school in time.To sum up,the main work and innovations of this paper are as follows:(1)Preprocessing various data of students in school and analyzing students' behavior.After extracting behavioral characteristics from various data sources,using K-means algorithm and isolated forest algorithm to identify students with abnormal consumption and performance changes,which helps the counselor to understand the individual situation of the students.(2)Proposing an algorithm for mining students' offline social networks,and visualizing offline social networks to identify isolated students.According to the idea of "canteen co-occurrence",that is,when two students often appear in the cafeteria at the same time,they may be good friends.Then using the consumption records of campus-card to mine students' offline social networks,and found isolated students.(3)Analyzing the correlation between student behavior characteristics and academic performance,and evaluating the results of academic performance prediction models based on traditional machine learning methods.We first use Apriori algorithm,Pearson correlation coefficient to analyze the influence of students' behavior characteristics on performance,then screen the students' behavior characteristics,and finally model the data and compare the prediction results of multiple groups of classifiers.What'more,Logic regression achieves the best performance,Precision,Recall,F1 and Accuracy are 0.75,0.74,0.74 and 0.85 respectively.(4)Proposing the Dis_GAT network,which operates the graph structure data,and transforming the academic performance prediction problem into the node classification problem in the graph structure.Firstly,introducing the concept of graph neural network,and transforming the students' numerical behavior into graph structure data,and then improving the Graphic Attention Network by introducing distance weight coefficients in addition to the attention mechanism weight,taking into account the influence of similarity between students' previous course scores on their later academic performance.The experimental results show that Dis_GAT network is more accurate than other methods.The evaluation indexes of Precision,Recall,F1 and Accuracy are 0.78,0.81,0.85 and 0.88 respectively,and the growth rate is greater than 8 percentage points.