Research On Academic Warning Method Based On College Students' Behaviors Analysis

In the big data era,educational data mining uses the theories and techniques of educational psychology,computer science and statistics to discover and solve various problems in educational research and teaching practice.The thesis aims to explore the factors that affect the performance of students based on the data of the six-year undergraduates in H-university,which includes 31597 students.The thesis assumes that students' graduation scores are related with the students' attributes and behavioral characters.A frequent sequential pattern mining algorithm based on the compression time slice sequence is proposed in order to find the top-k frequent behavioral traces.The algorithm is verified on the dataset of 27,939,974 e-card records about 2012 grade.Then,in order to verify the function of academic warning,the SVM algorithm is applied to the datasets,such as the behavioral trajectory and scores of college students in the first two academic years,to predict students' graduation scores.It provides new methods and ideas for the scientific and intelligent management of college students.The main work of this paper are as follows:1.Preprocessing data of the e-card and the score of each semester in the university.Based on the students' information of H-university from 2009 to 2014 grade,which includes the data of e-card usage records,course selection records,students' basic information and their score information,the thesis focuses on exploring the factors that affect college students' scores.Assuming that the scores of college students are inseparable from their behavioral trajectories.By integrating students' various behavioral,the students' coarse grained behavioral trajectories on campus can be recovered.Data preprocessing work concludes data cleaning,data integration,data transformation and data normalization.2.Proposing a new concept-law pattern according to the regularity of college students' behavioral data.The behavior of college students in school has periodicity and regularity.The thesis discovers the effect of regular behavioral patterns on students' score by analyzing the college students' regular living patterns.A new prediction model is created to verify the validity of the extracted features.3.A student's behavior trajectory can be separated into serval small time slices.Then a frequent sequential pattern mining algorithm(FSPC)based on the compressed time slice sequences is proposed.The students' coarse grained behavioral trajectories on campus can be recovered by integrating students' smart card usage,self-study records and the logs of entering/leaving dormitories.The most interested top-k behavioral trajectories will be found with the FSPC algorithm based on the compressed time slice sequences,which is helpful to discover the students' implicit behavior patterns.Furthermore similar behavior patterns among students can be identified.The experiments show that the FSPC algorithm has high efficiency and accuracy.4.Applying the above research results to academic warning.A student's final scores can be predicted according the front two years grades by the comprehensive analysis of the students' personal attribute,regular behavior patterns and behavioral trajectories.The problem students may be given academic warnings and corresponding behavioral suggestions.By conducting a large number of experiments based on real datasets,the FSPC algorithm is verified.The algorithm has improved the mining efficiency by using compression sequence and removing some unnecessary join operations.The findings of the thesis provide a new methods and data foundation that could be referenced by teaching management.