Predictive Model to Reduce Undergraduate Student Dropout at the Army Scientific and Technological Institute of Peru

Abstract

Student dropout represents a problem of great complexity and repercussion in the university educational environment, affecting students, academic institutions, and society. The Army Scientific and Technological Institute of Peru (ICTE) is not exempt from this problem, in addition to the small amount of historical data and limited access to information due to personal data protection issues. To address this problem, this paper presents a predictive model to reduce dropout in undergraduate students in ICTE. It uses Machine Learning techniques and compares its prediction levels through reliable performance metrics most used in the literature. The dataset initially concentrated information from only 72 students, classified into personal data, socioeconomic factors and academic performance. To overcome this challenge, we chose to generate synthetic data using the SMOTE technique based on the original dataset, thus facilitating the training of the classification algorithms, balancing minority classes and reducing biases in the prediction. The results obtained highlight the exceptional performance of the LightGBM model, which achieved 95% accuracy for training and testing. This model provides a strategic tool for ICTE to implement preventive measures to mitigate factors that negatively affect or hinder student retention in academic training. This approach promises to be beneficial for both students and the institution, contributing significantly to educational quality.