Clustering Approach for Modeling Course Difficulty Level in Adaptive Learning

The variations in students' abilities present challenges in achieving desired learning outcomes. Students exhibit different levels of comprehension regarding the complexity of learning materials. In this study, we aim to model the difficulty levels of learning materials using a clustering approach to offer personalized learning material recommendations within an adaptive learning framework. Data were collected from student scores of Data Structure course obtained between 2020 and 2022 at the University of Trunojoyo Madura. The dataset modeling process adheres to the CRISP-DM Methodology, employing the K-Means clustering method and Euclidean Distance. We employed the Within-Cluster Sum of Squares (WCSS) metric and silhouette score to identify the optimal K-value, which resulted in a value of 5, accompanied by a silhouette score of 0.629. Our analysis pinpointed Stack and Deque as the easiest, while Linked List emerged as the most challenging course material. Accordingly, we will assign a label of 1 to the easiest materials, 2 to those of medium difficulty, and 3 to the most challenging ones. These labels will serve as criteria for recommending suitable learning materials to students based on their knowledge level. We will assess the ratio between the overall knowledge level score and the difficulty level of the material. The ratio value will serve as a reference for determining material recommendations for students. Materials categorized with the highest difficulty levels are exclusively suggested to students demonstrating faster learning abilities with a high value of ratio, and vice versa. This approach is feasible for implementation in adaptive learning.