Predictive model based on machine learning for the early identification of student dropout in a public university

Resumen

Student dropout is a significant problem in public universities, with academic, economic, and social repercussions. This study aims to develop a predictive model based on machine learning techniques for the early identification of students at risk of dropping out. This will enable the implementation of measures to reduce dropout rates through intervention strategies such as personalized tutoring and socioeconomic support. The data used consists of academic, socioeconomic, and behavioral records from 2015 to 2024. The methodology employed is quantitative and predictive, utilizing algorithms linked to supervised learning techniques such as Logistic Regression, Decision Trees, Random Forest, and AdaBoost. The data were pretreated through cleaning, Z-score normalization, balancing using the SMOTE technique, and subdivision using the holdout technique into training (70%), validation (15%), and test (15%) subsets. k-fold cross-validation (k = 5) was applied during the training phase. The performance metrics considered to evaluate the proposed models were accuracy, precision, recall, F1-score, and the area under the ROC curve (AUC). The results obtained for each model indicated that the AdaBoost model performed best, achieving an accuracy of approximately 89% and an AUC of 0.957. Random Forest had an AUC of 0.902, and Logistic Regression had an AUC of 0.948. Therefore, it is concluded that the proposed model could be considered a suitable tool for the early detection of students at risk of dropping out.