Using data mining to understand student attrition in universities: A systematic review. (#1180)

Authors

Gonzales Requena, Ederick

Colupú Aquino, Eduardo

Abstract

Student dropout in universities is a complex phenomenon influenced by academic, socioeconomic, demographic and institutional factors. This systematic review aims to analyze how data mining has been applied to predict and mitigate this problem, identifying relevant patterns in academic and demographic data. A total of 117 original articles found in databases such as Scopus, Dialnet and Ebsco were evaluated, of which 30 met the inclusion criteria. The most commonly used techniques included Random Forest, decision trees and XGBoost, standing out for their high accuracy in predicting attrition. The most effective predictive models identified at-risk students with an accuracy of over 90%, allowing personalized strategies to be designed and institutional resources to be optimized. Thus, data mining is an effective tool for anticipating and mitigating dropout, but its impact can be maximized by integrating it with qualitative approaches that consider psychosocial factors, which would favor more inclusive and effective interventions.