Research And Implementation Of Prediction System Of College Students' Academic Performance Based On Web Logs

EDM(Educational Data Mining) technology is an interdisciplinary research which related to computer science, pedagogy, statistics and other disciplines. It is the extension and application of Data Mining technology in education. EDM has been a research hotspot in recent years when the collection of education related data becomes more efficient with the rapid popularization of Internet and information technology in education environment.Student performance prediction with high practical value and significance of academic research is one of the earliest and most popular applications of EDM. The data of existing research on performance prediction mostly come from the E-learning systems, intelligent tutoring systems or other special digital system, which lead strong pertinence but little universality. In order to draw a common performance prediction model, this thesis present a way that predicting students' academic performance by their daily Internet access record. This study testifies that the way to predict students' performance in some courses is feasible and has some practical value.The main work in this thesis can be divided into following four parts:First, the current situation of EDM technology is analyzed and the main content of the subject in this paper is described. Then some fundamental knowledge of Data Mining, the EDM, machine learning, NBC(Naive Bayes Classification) and LR(Logistic Regression) algorithm is introduced.Secondly, the data set and data preprocessing in this study are described. Data preprocessing is one of the most important steps before modeling of EDM, which is a key influence about the EDM results. The detail introduction about the source, characteristic, form and preprocessing of data set is demonstrated in this paper.Thirdly, the method that predicting students' academic performance from their access record on general websites is proposed. The experiment show that using six different data sets by NBC and LR algorithm to build a performance prediction model on Data Structure course.Finally, the conclusion shows that it is feasible that students' access record on general websites can be used to predict students' academic performance. Then this paper analyzes the deficiencies of the study, puts forward the improvement methods and references for the future work.This thesis demonstrated that there are some connections between access records on general websites and academic performance. More than 65% of unqualified students and 88% of passed students in Data Structure course can be identified. This study also finds that the student who watches more online videos and browses less technical websites is more likely to fail in the course. It is practical that applying the method in this research to practical system application for improving the at-risk students' achievement under teachers' helps.