Research On The Prediction Of Financial Aid For Students Based On Improved Gcforest

In order to solve the schooling problem of domestic students with financial difficulties,the government attaches great importance to the distribution of scholarships,and implements precise subsidies to poor students through national policies,school organizations and student supervision.However,due to the incomplete system and insufficient supervision,the effective distribution of scholarships has been hindered by many.The development of science and technology has enabled schools to effectively evaluate student information through big data,thereby effectively assisting the precise granting of grants.In this context,this thesis analyzes and processes the multi-source data of students in school,and builds a predictive model of student aided payment based on the deep forest algorithm.The main work of this thesis is as follows:(1)Analyze the acquired multi-source information data of students in school,and use data preprocessing,feature construction,and feature selection based on correlation analysis to ensure the scientific validity of the data.Using the multi-view information fusion method,the student's multi-dimensional data at school is divided into two different views in two ways,and the two views are merged using Canonical Correlation Analysis(CCA)to strengthen the relationship between attributes.(2)Improve the way the gc Forest algorithm handles class imbalanced data,using the SMOTE algorithm,C?SMOTE algorithm and increasing category weights to process the data,and propose the method of processing class imbalances to be integrated into the gc Forest algorithm framework,And built SMOTE+gc Forest,C?SMOTE+gc Forest,and weight+gc Forest models are built to predict the student's award.(3)Improve the reorganization method of the feature vector of the gc Forest algorithm and the generation method of each forest class vector in the cascaded forest.The CW?gc Forest algorithm is proposed to add the calculation of the similarity of the vector and the calculation of the model performance during the k-fold cross-validation to the gc Forest algorithm,and the C?SMOTE+CW?gc Forest model is built to predict the student's award.The final comparison experiment results show that the performance of the algorithm model proposed in this thesis has been improved to a certain extent on the student information data set,which verifies the necessity of considering the relationship between attributes and the effectiveness of the CW?gc Forest algorithm proposed in this thesis,as well as the pairing of C?SMOTE+CW?gc Forest model.The validity of the forecast of grants can provide a practical basis for the accurate issuance of grants.