Application of clustering algorithms in the identification of students at risk of dropping out in higher education: a systematic mapping review

Student dropout in higher education is a global challenge that affects educational quality, institutional performance indicators, and students' academic development. With the increasing availability of educational data and advances in data mining techniques, unsupervised learning methods such as clustering have emerged as promising approaches for identifying dropout risk patterns without the need for labeled data. This study aimed to systematically map and analyze clustering techniques applied to segment students at risk of dropout in higher education between January 2010 and July 2025. To this end, a Systematic Literature Mapping was conducted following the PRISMA protocol, which included the stages of search, screening, data extraction, and thematic analysis of the studies. The results indicate that algorithms such as K-Means, DBSCAN, HDBSCAN, BIRCH, SOM and hierarchical methods have been used with varying levels of effectiveness, with internal validation metrics predominating. Academic and behavioral variables were the most frequently used in cluster formation, while psychosocial variables, although still limited, demonstrated significant potential. The findings show that clustering is a promising approach to support student retention strategies, although further progress depends on expanding datasets, strengthening validation methods, and integrating results into institutional policies designed to promote student persistence and academic success.