Identifying Those Who Really Need Subsidies: A Data-Driven Approach

Subsidies and supportive policies have been often offered by universities to students with financial difficulties. However, it is a non-trivial task to scientifically and accurately identify students who really need subsidies with the traditional apply-review method. In a smart campus, students use a campus card to spend money on affairs such as eating in the canteen, taking a bath, and shopping in the supermarket for their daily living, and the consumption records potentially reflect the economic level and the living habits of students. To this end, we herein propose a data-driven approach that combines statistical methods and machine learning models (CSML) to accurately identify students who really need financial aid. CSML first preprocesses the consumption data and extracts seven informative features that are closely related to eating and bath charges. Second, the overall consumption portraits of different gender, grade, and financial difficulty levels are obtained, and false poverty and suspected poverty students are excluded from the study based on the average consumption. Third, a supervised classification model is used to predict the type of financial difficulties a student belongs to, followed by a statistical method to check whether the student predicted with financial difficulties spends more. Experiment results show that CSML achieves a 96% precision in identifying students with financial difficulties, which reveals the power of CSML in helping evaluate the effect of subsidies.