<![CDATA[Purpose – This study compares the K-Means and Fuzzy C-Means (FCM) algorithms for mapping student academic risk using in-course academic performance data. Methods – The dataset consisted of 35 students and included assignment scores, quiz scores, midterm examination scores, attendance, learning participation, employment status, and language variables. The data were preprocessed through cleaning, identity anonymization, and min-max normalization to ensure that all attributes were measured on a comparable scale. The experiments were conducted under two clustering scenarios, namely K=2 and K=3. Findings – In the K=2 scenario, both methods produced the same separation between low-risk and high-risk student groups. After the clustering results were mapped to the actual Pass/Fail labels using a majority-vote approach, 27 students who passed and 7 students who failed were correctly identified, with no false positives and 1 false negative. These results yielded 97.14% accuracy, 100% precision, 96.43% recall, and a 98.18% F1-score. In the K=3 scenario, K-Means formed three distinct groups containing 27, 4, and 4 students, whereas FCM produced a more gradual distribution of 13, 14, and 8 students. Research implications – These findings indicate that K-Means is suitable as a fast baseline for binary risk screening, whereas FCM is more informative for gradual risk interpretation in academic early warning systems. Originality – This study contributes by showing the different practical value of hard and soft clustering for identifying clearly at-risk and borderline students using routinely available in-course academic indicators.]]>
Copyrights © 2026
