Research On Variable Membership 2C-FSVM And Its Application On Assisted Recognition Of Poverty Students

Recognition of poverty students is one of important issues in China's colleges,which plays a crucial role in the fair assignment of resources.Currently,big data and machine learning offer an effective approach to addressing this issue.Considering such problems as the imbalanced data due to the low proportion of poverty students and different interference factors that result in false identification,this thesis proposes a variable membership 2C-FSVM(Dual-C-parameter Fuzzy Support Vector Machine)based recognition method of poverty students.The variable membership 2C-FSVM method not only addresses the uncertainty and imbalance of real-life datasets of poverty students,but also uses different fuzzy membership calculation ways depending on the distance of a sample point from the center of one class,which characterizes the uncertainty of sample points more precisely.Based on the variable membership 2C-FSVM,we develop a method to assist the recognition of poverty students,which not only provides a more objective and accurate recognition of poverty students,but also helps to discover the suspected “hidden poverty” and “false identification” students,thus improving the efficiency and accuracy of the recognition of poverty students and promoting the targeted support in poverty alleviation.First of all,this thesis investigates related work on the assisted recognition of poverty students and SVM,by focusing on the imbalanced and uncertain issues during the SVM research.Secondly,this thesis introduces related techniques that will be used,including SVM,random forest based feature selection,evaluation methods of imbalanced datasets,and the K-fold cross validation method.Thirdly,aiming at the existing problems of the recognition of poverty students,this thesis proposes an improved variable membership 2C-FSVM model,and derives its Lagrange duality problem.In this improved model,each data point is not only given a different penalty based on its class,but also is calculated with different fuzzy memberships based on its distance from the center of the class.Moreover,the proposed variable membership 2C-FSVM method is compared with other SVMs on some public imbalanced datasets.The experimental results show that the variable membership 2C-FSVM is less sensitive to the imbalanced and uncertain datasets,and the recognition of the minority class in uncertain datasets is more accurate.Finally,this thesis applies the proposed variable membership 2C-FSVM model to assist the recognition of poverty students in a college.A complete procedure for this application is given,and then with campus card data,the variable membership 2C-FSVM is trained and tested,and a recognition model of poverty students is obtained.This model is then used to assist the recognition of the poverty students,and to discover the suspected “hidden poverty” and “false identification” students,thus achieving the recognition of poverty students in colleges.