Research And Application Of Performance Prediction Model Based On Data Mining

Educational Data Mining（EDM）is an important field in Data analysis.With the new environment of network education,EDM has gradually become a rich research field in computer science in recent years.EDM focuses on exploring he internal hidden information in education data,which is used to predict the status of students,provide timely guidance,help teachers and administrators to scientifically and effectively obtain the past trends and advantages and disadvantages of the status of students in learning.Erformance prediction is an important direction of EDM research.For this field,researchers start from existing data analysis algorithms（SVM,C4.5 decision tree,XGBoost,naive Bayes and other algorithms）,gradually improve and optimize,and constantly put forward new ways to solve problems.However,Previous studies did not take into account the difference in the influence of different characteristics on students’ learning of different subjects.Therefore,in order to solve the above problems and improve the accuracy of prediction results,this paper elaborated two improvement ideas and optimized the prediction method and the performance of the model.The main research contents are as follows:（1）Now every school has students exists the phenomenon of dine,and currently a research scholar in the field of EDM in student performance prediction,tend to ignore the problem,the default characteristics of the influence degree of the students to learn all the subjects are the same,does not take into account under the condition of the same background,dine in why the students’ academic performance,Therefore,this study firstly classifies subjects to solve the problem that each feature has different influence on students’ learning of different subjects.By sorting the features of each subject and analyzing the importance of their features,teachers of each subject can timely understand which groups of students need more guidance,which is of great help to the design of high-quality teaching plans.By analyzing the correlation between the importance of features of different subjects,it is found that in the same subject category,the same feature has similar influence on these subjects,indicating that the subjects of the same subject category have a high degree of correlation,so students’ scores are finally predicted according to the subject category.（2）In this paper,a hybrid prediction model is proposed based on XGBoost algorithm,and a hybrid feature selection model combined with XGBoost is proposed,which is named HFS-Xg Boost.Hybrid feature selection combines the CFS algorithm in the filtering method and Lasso algorithm in the embedding method.Since the advantages and disadvantages of the filtering method and the embedding method on the weight coefficient are different,the combination of the two methods can fully promote the advantages,avoid the disadvantages and improve the performance of the model.