Behaviors and features selection of online learning data

To identify any patterns in learning behaviors, learners' learning behavior data are captured and stored in many online learning platforms. It is crucial to determine which behaviors and which features of a behavior are most related to the specific analytics task. To do this, feature selection method has often been applied to determine a global reduced feature space. However, little attention has been paid to select the behaviors and features within behavior simultaneously. In this work, we propose a two-level feature selection method which can determine the importance of behaviors and features simultaneously. The proposed method is embedded into the classical k-means to cluster a famous e-learning dataset. Our experimental results show that the proposed method is an effective way to improve the clustering performance significantly.