Enhancing Student Performance LMS Data with Feature Engineering for Improved Cluster Quality

The rise of Learning Management Systems (LMS) in educational environments has generated vast amounts of data, offering valuable insights into student behavior and learning outcomes. However, the raw data from LMS platforms often suffer from noise and lack of structure, hindering effective analysis. This paper explores the application of feature engineering to LMS data, aiming to enhance the quality of clustering and improve predictive performance. LMS data from a total of $\mathbf{6 4, 2 3 1}$ students’ Moodle logs were collected from a private university in Malaysia in January to April 2019 for analyzing students’ behavior from 5 undergraduate Computing courses. By systematically transforming and augmenting raw data, new informative features are created to capture key aspects of learner interactions. Our approach is validated through extensive experiments using four clustering techniques such as KMeans, Agglomerative, BIRCH, and DBSCAN. The results reveal a 35.15% increase in the Silhouette Score, 130.82% Calinski-Harabasz Index, and 39.59% decrease in the Davies-Bouldin index, demonstrating that engineered features and outliers removal significantly improve the coherence of clustering results. The findings underscore the importance of feature engineering in Educational Data Mining.