Study On Accurate Prediction Of University Subsidy On Multi-Strategy Method

In recent years with the popularization of higher education,The number of students in colleges and universities has also increased rapidly.At the same time,for the economically disadvantaged students in the university,the party and the government also issued a series of documents to help the students in difficulty,so that they can successfully complete their studies.At present,colleges and universities decide whether to subsidize poor students and determine the level of student financial aids according to the “student application—teacher signature—school approval” procedure.Such procedures are inevitably affected by some human factors.At the same time,with the advent of big data and the Internet era,it provides new ideas and technical support for universities to deal with the issue of precision funding.This thesis uses students from a university in Sichuan Province from 2013 to 2015 as a data mining object,and using data mining technology and machine learning algorithm to establish a precise prediction model for college bursary and use F1 value as the evaluation index of the model.This paper first performs data preprocessing,followed by exploratory data analysis and feature engineering to obtain excellent feature data.Then the accuracy of the model is verified by cross-validation method,and the effects of random forest algorithm,AdaBoost algorithm,support vector machine and GBDT algorithm on the bursor precision prediction model are compared in the model construction process.It is found that the GBDT algorithm is relatively optimal.In order to further improve the performance of the bursary prediction model,this paper uses the Stacking method in ensemble learning to perform model fusion.Finally,combined with the GBDT algorithm,the AdaBoost algorithm and the random forest algorithm are further combined to enhance the performance of the model.This multi-strategy based combination model can help managers to obtain daily behavioral data of college students during school and the level of funding for poor students in the model.Thus assisting in the discovery of “falsely identified” and “invisible poverty” students.It has practical significance for optimizing the evaluation and subsidies of college bursaries,saving the time of relevant managers of colleges and universities,and improving the supervision mechanism for student subsidies.