The Application Of Data Mining In The Prediction Of College Students' Grade Repetition

Under the background of Web2.0 era,information technology has been rapidly innovated,and massive data accumulation has made data mining technology develop unprecedentedly.The application based on data mining has extended to many fields,among which educational data mining is an important branch.How to wisely use data mining technology to solve the problems existing in student management and education and teaching in colleges or universities has become the main research in the educational field.Therefore,using data mining technology to build a reasonable prediction system for students who will repeat a grade plays a vital role in the reform of college teaching,which is of great academic value and practical guiding significance.The main work of this paper is as follows:Three types of data,including students' historical scores,students' online diaries and admission basic information,were selected for analysis and research.After some pre-processing work,such as data cleaning,denoising and conversion,the students' historical scores table,students' network log table and students' basic information table were sorted out,and the build the correlation between the research in the table and grade repetition.The results showed that the correlation in grade repetition has higher relationship with the number of failures,failure rate and failure credits in the historical scores,and the correlation between the network log behavior data and the basic admission information of students and the retention of students was generally lower,while the correlation between the college entrance examination scores and the retention of students in the basic admission information was higher.According to the three types of data of freshmen's historical scores,students' weblogs and basic admission information,it is predicted whether there will be grade failure in the next2-3 years.Four prediction models based on Naive Bayes,logistic regression,decision tree and BP neural network were constructed.First,only historical grade data were used to predict grade repetition,then network behavior data and basic admission information were added,and a variety of different feature combinations were selected to predict grade repetition.Three evaluation indexes,precision,recall and F1-measure,were used to evaluate the prediction results of different feature combinations of different prediction models.The results show that the BP neural network is superior to the other three models with the accuracy of 71% and the recall rate of 84%.Further combined with feature selection and screening different featuresfor prediction,the prediction ability of the model is further improved,among which BP network has the best prediction result,with an accuracy rate of 83% and a recall rate of 90%.With the expansion of data sources,the prediction effect of BP neural network presents a downward trend.A hybrid GABP prediction model based on genetic algorithm and BP neural network is constructed,in the consideration of the further optimization on the prediction results used to improve BP neural network algorithm,which tend to fall into local optimum.The results showed that the accuracy rate of GABP model reached to 82%,and the comprehensive evaluation accuracy rate and recall rate index value reached to 86%,which could effectively predict the students with grade failure risk.