The Discovery Of Financially Incompetent Students In University Based On Behavioral Data

The poverty alleviation has always been the focus of all kind of universities.There is a lot of difficulties in this work,which has a long process,involves lots of members from several departments.The traditional way of poverty alleviation is very inefficient.The goal of this thesis is to use a variety of behavioral data generated by students in the campus to find a reliable method of determining the economic level of students,and to apply this method for actual business.Finally,our goal is to establish a complete Student Economic Level Assessment System,which provide support and guidance for the relevant staff of poverty alleviation in universities.In general,at the start,this thesis deals with the process of data integration and cleaning,then with the extraction and selection of features,the construction and analysis of algorithm model,and finally with the whole process of system design and implementation.Among all these works,in the data integration and cleaning part,initially,we established a standard database for university according to the relevant national standards.Secondly,we analyzed and cleaned all data problem,according to the characteristics of different data source systems,focusing on the completion of missing field from different data sources.In the feature extraction and selection part,as the universities runs,we set up different time period: days,weeks,months and semesters,and extract the timing series according to different time periods.The extracted features are divided into basic statistical features and complex features.After the feature extraction completed,the post-pruning C4.5 decision tree is used to screen the features.In the construction and analysis of the algorithm model,we use RNN to construct the model according to the timing characteristics of data and features.We combined advantages and disadvantages of the two most commonly used RNN methods,LSTM and CW-RNN,and described how it work and the result we got did.Finally,in the system design and implementation stage,as the online process is completed,we put forward the concept of dynamic management of poverty alleviation based on the results of the algorithm model,in order to improve the poverty alleviation work of the university from the annual cyclical work to a real-time care.The main achievements of this thesis are as follows: 1)The establishment of university data standards,the integration of student data onto an unprecedented large-scale,combined with the use of the actual systems,targeted to complete the analysis and cleaning of data;2)Explored a series of characteristics related to the economic situation of students;3)Proposed an improved recurrent neural network model to deal with the characteristics above,and achieved a good result;4)Combine the actual implementation of the poverty alleviation in universities,we developed a system for the certification of economically difficult students.The results of the final algorithm validation and the use of the system showed that the Poverty Index has a good reference value of the assessment of the student's economic level.It also shows that the use of big data analytics in university is feasible and has a practical value.