Accuracy of Learning Method Implementation in Higher Education Using K-Means Clustering and Silhouette Coefficient

Abstract

Background: The Covid-19 pandemic significantly transformed the learning process in higher education, forcing institutions to quickly adapt to unprecedented challenges. Traditional face-to-face learning was no longer feasible due to health restrictions, and this condition accelerated the integration of technology into teaching and learning activities. As a result, online, offline, and hybrid learning methods emerged as the primary alternatives for sustaining academic activities. However, the rapid shift also highlighted a critical issue: the effectiveness of these learning methods varied widely depending on institutional readiness, available resources, and student adaptability. Determining the most effective method has therefore become essential to ensure quality outcomes, maintain student performance, and support the continuity of higher education in the post-pandemic era.

Aims: This study aims to identify the appropriate post-pandemic learning strategy in higher education by applying the CRISP-DM methodology and the k-means clustering algorithm.

Methods: The dataset consists of 65,778 student records collected from 2015–2020, preprocessed through data reduction, cleaning, and transformation. K-means clustering was applied using Orange, an open-source data mining tool, and evaluated with the Silhouette Coefficient.

Result: The results show that offline learning produced the highest total frequency, hybrid learning was in the medium range, and online learning the lowest. Silhouette Coefficient scores indicated cluster quality in the medium structure category, with values of 0.47, 0.56, and 0.65 across three clusters. These findings suggest that offline learning remains the most effective method under normal conditions, hybrid learning is more suitable during pandemic or transitional periods, while online learning can serve as an alternative depending on institutional or governmental policies. The study concludes that clustering-based analysis provides practical insights for designing adaptive, data-driven learning strategies in higher education.