Clustering-Based Student Segmentation for Adaptive Learning Strategies
Year: 2026
Type: conference-paper
Abstract: The majority of schools and colleges treat a classroom of students as though they are practically the same, but anyone who has taught knows that this is not the case. Students arrive with varied study habits, homes, motivation levels and academic backgrounds. When they are treated the same way, the same issues can be expected: some of them become bored, others lag behind, and the percentage of those who simply tune out without anyone noticing is high. This paper presents a segmentation analysis where an unsupervised clustering algorithm was applied to divide a dataset of 6,607 students into groups by 19 behavioral and socioeconomic variables and then each group was analyzed to determine what type of teaching support would actually be appropriate to that group of students. K-Means and Ward-linkage Agglomerative clustering were both run on the Student Performance Factors dataset after ordinal encoding and standardization. The cluster count was selected using four internal indices: Elbow, Silhouette, Davies-Bouldin and Calinski-Harabasz. The two algorithms converged on two groups, with a Silhouette score of 0.160 and an Adjusted Rand Index of 1.0 across methods. Internet access at home was found to explain 91% of the feature signal between the two groups using a Random Forest surrogate and SHAP attribution. In addition to the technical divide, this paper pairs each cluster with specific intervention recommendations, verifies the demographic composition of each cluster on the issue of fairness, and indicates the proportion of students in the 30th percentile or lower within each cluster. A number of gaps in the research of educational data mining are addressed: the tendency to halt at visualization, the inability to explain cluster-label assignments, and the absence of equity checks in segmentation pipelines.
Institution: Jesselton University College
Language: English
Subfield: Computer Science Applications
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7214770828