Application of Non-Supervised Learning Tools and Visualization Techniques to Understand the Intrinsic Freshmen Enrollment Segmentation: An application to freshmen engineering students.

Abstract

Identifying significant subgroups among first-year students is crucial for designing educational policies that foster their academic and personal development. This study presents a data-driven methodology to segment first-year students using sociodemographic factors, admission test scores, and initial academic performance. Data from 7,866 engineering students enrolled at a Chilean university between 2005 and 2017 were analyzed. By applying Self-Organizing Maps (SOM) in combination with the k-means algorithm, our methodological approach enables the visualization and classification of complex student profiles. SOM provides a two-dimensional representation of the data, while k−means refines the generated clusters, offering a more coherent perspective of intrinsic segmentation. The results provide a robust framework for higher education institutions to develop targeted policies and strategies tailored to the characteristics and needs of different student groups. Although focused on a Chilean context, the proposed methodological approach holds broad applicability across various educational institutions, contributing to the development of evidence-based policies that promote academic progress and educational equity.

Keywords

• Freshmen enrollment analysis
• Data-driven methodology
• Academic information visualization
• Higher education institutions