University Stipend Prediction Based On Multi-view And Classification Ensemble

With the educational informatization developing,data-driven education reform has gradually become a research focus.It is the arriving digital campus era that has brought new opportunities for universities to promote convenient and efficient stipend prediction.The implementation of digital campus allows the data such as living habits,learning behaviors,and test information of students to be easily obtained and used,and analysis of these data is of great significance for improving education and teaching.It is a great importance for universities to do successful stipend assessment.The information for traditional stipend assessment is limited.The student’s financial situation is closely related to the student’s school behavior,so the student’s school behavior data can be used to predict the students who need stipend quickly and efficiently.The behavior data of students at school is from different sources.If these data are simply combined,the association and complementary information between data cannot be fully utilized.The multi-view feature learning methods can take into account the mutual connection between different views and reduce the amount of redundant information,which helps to improve the classification effect.The classification ensemble methods can obtain better results than a single classifier,which can further improve the prediction results.Therefore,a multi-view combined with classification ensemble model for university students’ stipend prediction is proposed.This model mainly includes four steps: firstly,the students’ multi-dimensional data in the school is constructed into two different views according to learning performance and life behavior,and then used.The Multi-view Uncorrelated Discriminant Analysis(MULDA)method processes the different datasets and obtains more discriminative features.The third step is multi-view information fusion.Finally,the Gradient Boosting Decision Tree(GBDT)ensemble method is used to predict the stipend labels of students.The experiments show that the multi-source data from universities,combined with the multi-view method and classification ensemble method,gets better results than the single-view and single-classifier methods.The existing discriminative canonical correlation analysis algorithms for two views do not consider the comprehensive factors of the single view’s classes correlation and the discriminability of the view combination features.Therefore,an Enhance-Discriminative Canonical Correlations Analysis(EN-DCCA)method is proposed.The optimization goal of this method is to minimize intra-class correlation while minimizing inter-class correlation of views,and take into account the discriminability of combined features of different views.So EN-DCCA further strengthens the discriminability of the features,and combines the RKN-CE classification ensemble method to construct the prediction model.Experiments show that the proposed model gets better results than the single-view and single-classifier methods.