Shanghai Open University
China
The Open University of Hong Kong,
Hong Kong
With the rise of online learning and the growing number of online learning data and student data which are stored in the LMS and CMS, many researchers are carrying out online learning behaviour analysis and student performance prediction. But, in recent years, more and more teachers and students prefer to use mobile smart terminals and social applications to interact with each other and discuss learning problems rather than traditional LMS or CMS. Those data are difficult to collect, and so researchers have much fewer effective data available from online learning platforms than before. However, we can still conduct learning behaviour analysis and predict students’ course grades using machine learning tools based on limited data. The purpose of this paper is to carry out correlation analysis between some student characteristics and features of online learning behaviour and course grades, and to attempt to build an effective prediction model based on limited data.
The prediction label in this paper is the course grades of students, and the eigenvalues available are students’ age, gender, connection time, hits count, and days of access. The machine learning model used in this article is the classical three-layer feedforward neural network, and the scaled conjugate gradient algorithm is adopted. Pearson’s correlation analysis method is used to find the relationships between course grade and the student eigenvalues.
The days of access had the highest correlation with course grades, followed by hits count, and connection time was less relevant to students’ course grades. Student age and gender had the lowest correlations with course grades. Binary classification models have much higher prediction accuracy than multi-class classification models. Data normalization and data discretization can effectively improve the accuracy of prediction of machine learning models, such as the ANN model used in this paper.
This article may help teachers to find some clues for identifying students with learning difficulties in advance, and give timely help through the online learning behaviour data. This research showed that acceptable prediction models based on machine learning can be built using a small and limited dataset, and data preprocessing is important for building a better prediction model. However, introducing external data, especially learning and social data, in mobile devices into machine learning models to improve their prediction accuracy is still a valuable and hard issue.