Cluster Analysis of Student Learning Process on Digital Learning Media Based on Visual Intuition Using Adaptive Moving Self-Organizing Maps

A clustering technique that utilizes students' learning activity log data is one of the many methods to mine information regarding students' behavior during the learning process. We group students using Adaptive Moving Self-Organizing Maps (AMSOM) and analyze the result to reveal students' learning patterns as an effort to give better and more relevant learning feedback to them. The research starts with data collection process from students' learning activity log data that is gathered while using the digital learning media, converting the sequential students' activity data into features, grouping the data objects using AMSOM, evaluating the clustering result using Quantization Error (QE) and Topographic Error (TE), determining the level and tasks to be analyzed, and finally analyzing the students based on the visualization from the clustering results and linking them with influential values and features. We compare the result of AMSOM and traditional Self-Organizing Maps (SOM) on 12 tasks in level 5 and discovered that AMSON improves clustering performance by reducing the average error rate by 96.33% (from 0.1556 to 0.0057) for QE and 98.18% (from 0.1666 to 0.0030) for TE. A deeper analysis is then carried out on three first tasks from level 5. The number of clusters from the first task is 28 clusters which indicates that students have many different patterns of learning strategies. While working on the second and third tasks, the number of clusters decreased become 18 clusters and 17 clusters, respectively. This result shows that students' learning strategies have more similar patterns to reach the correct answer. The clustering changes among tasks indicate that students change their thinking strategy and increase the degree of understanding to solve the problems.