Performance Pattern Mining for Higher Education Students in Blended Learning Using Clustering Algorithms

Analyzing student learning pattern will help to infer useful insight from educational datasets and improve the learning process. These patterns will be helpful in identifying areas of the subject/course that need some changes, students that need more attention, and predictive analysis of their performance. To achieve this goal, it is essential to assess the level and growth of students in every scenario. The digitization of the learning process also leads us to a variety of data, including interactive features, prior knowledge of the subject, exam grades, earlier grades, academic detail grades, and features related to their knowledge level. This data should be processed through appropriate methods to analyze student performance. The analysis of question levels is another major focus of this paper, as question difficulty has an impact on students’ performance levels as well. This study dataset is of IT stream MTech(Master of Technology(Grp-1) and MCA(Master of Computer Application(Grp-2)), Integrated courses students. Both courses have blended learning modes, including face-to-face instruction with online learning activities. The Gnomio Moodle platform was used to assess their quiz performance and activity. The Statistical analysis of data has been done by comparing the facility index and discrimination efficiency of each item/question. Along with this their quiz score, board scores, CGPA, forum posts, prior knowledge, topics missed in class, etc. features were considered to analyze their performance and infer learning patterns through the K-means, Hierarchical, and GMM(Gaussian Mixture Model) clustering methods. All these 3 cluster methods were applied to both groups. To test the accuracy of clustering models, Internal validation matrices Silhouette score was used. The k-mean and GMM clusters effectively group students according to their learning patterns. Both have nearly the same Silhouette score of 0.303 and 0.313 respectively for the GRP-2 dataset.