Key Course Selection for Academic Early Warning Based on Gaussian Processes

Abstract

Academic early warning (AEW) is very popular in many colleges and universities, which is to warn students who have very poor grades. The warning strategies are often made according to some simple statistical methods. The existing AEW system can only warn students, and it does not make any other analysis for academic data, such as the importance of courses. It is significant to discover useful information implicit in data by some machine learning methods, since the hidden information is probably ignored by the simple statistical methods. In this paper, we use the Gaussian process regression (GPR) model to select key courses which should be paid more attention to. Specifically, an automatic relevance determination (ARD) kernel is employed in the GPR model. The length-scales in the ARD kernel as hyperparameters can be learned through the model selection procedure. The importance of different courses can be measured by these corresponding length-scales. We conduct experiments on real-world data. The experimental results show that our approaches can make reasonable analysis for academic data.