Feature Selection for Clustering Online Learners

As one of the most important approaches for learning analytics and educational data mining, various clustering algorithms have been explored and compared in the analysis of online learners by their behavior. However, choosing which features for clustering has a strong impact on the quality of clustering, and has not received enough attention yet. By using an entropy-based feature selection method, this research broadens the range of candidate initial features and provides efficient algorithms for extracting the most important features. Experiment with real-life data reveals that this method not only overcomes the data sparsity and complexity problems for clustering in high-dimensional feature space but also surpasses dimensionality reduction methods like PCA in interpretability.