Poor Performance Discovery Of College Students Based On Behavior Pattern

Academic performance of college students is the main concern of educational institutions. Early discovery of students with poor performance is valuable and essential. According to common sense, the differences between college students performance mainly depends on how hard they work, which can be reflected by their living and behavior habits. With the rapid construction of digital campus, university as the main range for students' life can not only serve convenience but also save daily records. The popular using of smart card makes it easy to outline students behavior pattern with rich data. The purpose of our work is to discover college students with poor performance based on their behavior pattern and analyze the correlation between students behavior and their performance. In this paper, we propose a general framework to discover students with poor performance. Firstly, we preprocess the raw behavior data recorded by smart card and describe students behavior pattern by extracting behavior features in two perspectives including statistics and relevance. Then we employ a multi-task model to learn performance of every course simultaneously. Our experiments on a real world data set of college students show a good outcome. We do a further analysis on what behaviors related more to academic performance. Moreover, our experiments indicate that our framework is feasible to early warning with semi-semester behavior features.