Supporting poverty-stricken college students in smart campus

摘要

Chinese colleges have formulated supporting polices to help poverty-stricken college students to deal with the barriers in their living and learning. The difficulty in fully collecting the required information related to student's financial status and the imbalanced-data classification problem caused by the small proportion of poverty-stricken students among total students makes it a challenging problem. This problem results in a heavy workload for the college staff to identify poverty-stricken students, determine the amount of corresponding subsidy, and execute the supporting polices in an efficient way. Therefore, this paper attempts to address the above-mentioned challenges by proposing a smart campus system, which makes use of campus big data to identify poverty-stricken students and support the decision-making on the subsidy for them. The proposed system can also alert the counselors to . provide psychological support for students in trouble. The major contributions of this research are as follows. Firstly, in addition to the features of students' amount of consumption on campus and its statistical characteristics used in existing researches, this paper proposes new features that describe diversity of consumable commodities, preference of consumption location and price, and characteristics of students' campus activities. Secondly, in order to solve the problem of dataset imbalance, four imbalanced data processing methods (Subsampling, Resampling, Cost-sensitive learning and SMOTE) have been applied to produce four different experimental datasets, and five classification algorithms (Random Forest, J48, Naieve Bayes, SMO, Logistic regression) have been used to train the classification model on each dataset. The experimental results indicate that the model based on Resampling and Random Forest achieves the best performance in F1-measure of poverty-stricken students, among the combinations of four imbalanced processing methods and five classification algorithms. In addition, a method of quantization of subsidies, and strategies of early warning and counseling for students are also described in this paper. A system was developed based on the above-mentioned methods, which meets the needs of individualized and diversified support for poverty-stricken students. The methods and the proposed system have been put into practice, and it is serving more than 17,000 students. The system has significantly improved the efficiency and quality of student management, and reduced the workload of college staff.