食堂消费大数据可以精准识别贫困生吗?——基于本科生行为数据、行政数据和问卷数据的实证研究

Can Canteen Consumption Big Data Accurately Identify Students with Financial Difficulties?——An Empirical Study Based on College Students’ Behavioral Data, Administrative Data and Survey Data

基于某“双一流”建设大学本科生食堂消费大数据、助学金发放的行政数据以及家庭基本信息的调查数据,利用统计学习模型对基于食堂消费大数据识别家庭经济困难学生的精准性进行了估计。研究发现:食堂消费大数据对家庭经济困难学生的识别精准率仅能达到约60%,采用更精细的消费时间序列数据可以将识别精准率提高到约65%,进一步结合家庭基本信息的问卷调查数据则可以将识别精准率提高到约92%。相比传统的逻辑回归,采用提升树和支持向量机等判别模型可以提高模型对家庭经济困难学生的识别能力。本文的研究发现说明,仅仅利用食堂消费数据作为补助依据的精准度仍有待提高,应将食堂消费数据与学生家庭基本信息数据相结合来提升资助精准度。

Based on the big data of undergraduates’ canteen consumption, the administrative data of financial aid and the survey data of socioeconomics information at a double first-class university in China, the paper examines the accuracy of identifying students with financial difficulties by using the statistical learning models and draws the following conclusions: the accuracy of identifying those undergraduates with financial difficulties is 60% by using big data of canteen consumption;and the accuracy can be improved to about 65% by using more refined consumption time series data;the accuracy can be further improved to about 92% by combining consumption data with questionnaire survey data of socioeconomics information.Compared with traditional logistic regressions the accuracy of discriminant models such as Lifting Tree and Support Vector Machine is significantly higher.These findings indicate that the accuracy of identifying students with financial difficulties needs to be improved by combining canteen consumption data with students’ socioeconomics information.