Data mining for predicting university student dropout: Literature review and experimentation

Abstract

The university dropout rate represents a critical challenge for higher education institutions due to its economic, academic, and social impact, not only on students but also on the institutions themselves. This study adopts a mixed approach, consisting of a systematic literature review and a practical experimentation using data mining techniques, considering that in recent years, data mining has become a powerful tool for predicting this phenomenon through the analysis of large volumes of data. The objective is to identify the determining factors associated with university student dropout and to evaluate the effectiveness of different machine learning algorithms for its prediction. The systematic review was conducted following the PRISMA method and included publications from the last five years in indexed databases. Subsequently, a predictive experiment was carried out using a dataset with 34 variables and more than 4,000 records, applying supervised models such as decision trees, Random Forest, and Gradient Boosting. The results show that economic factors, academic performance, and institutional conditions are the main predictors of dropout. In the experimental phase, Random Forest and Gradient Boosting yielded the best predictive performance, while decision trees offered high interpretability useful for institutional decision-making. The study concludes that the integration of data mining and machine learning models enables early identification of at-risk students and strengthens retention strategies in higher education.