Research on Construction of Student Academic Early Warning Model Based on Ensemble Learning

The educational data mining technology could support teaching management and it has received more and more attention. This paper focuses on the prediction problem of student academic performance and constructs the ensemble learning based early academic warning model to make accurate prediction. This paper collects students’ behavioural data from different campus systems, i.e., teaching system, online learning system, education and research system, library system and campus e-card consuming system, and make standardization processing for the original data. By utilizing the educational big data, this paper constructs an ensemble learning based early academic warning framework to predict students’ academic performance. The framework contains three modules, Correlation analysis and Potential Features Selection, Data Reduction and Ensemble Learning based Academic Performance Prediction. Correlation analysis and Potential Features Selection module first analyses the correlation relationship between independent variables (i.e., indicators) and dependent variable (i.e., students’ academic performance) through Apriori algorithm and then selects indicators which have strong correlations with students’ academic performance as potential features. In order to identify accurate features from potential features, the PCA method is utilized for Data Reduction. In the Ensemble Learning based Academic Performance Prediction module, an ensemble learning-based model is constructed to make accurate academic performance prediction, which aggregates Random Forest Classifier, XGBoost Classifier, BP Neural Network as the base classifiers and takes multinomial logistic regression classifier as the meta classifier. The prediction results could provide support for educational administrators and teachers, and give early warning for students.