Data Science Framework for Evaluation of Entrance Exams and Identification of Academic Patterns

Huertas-Condori L.N.; Cabana-Yupanqui S.B.; Chaparro-Cruz I.N.; Chaparro-Guerra A. · Proceedings of the 2025 IEEE 32nd International Conference on Electronics, Electrical Engineering and Computing, INTERCON 2025 (2025) · DOI: 10.1109/intercon67304.2025.11244670

Resumen

This study proposes a machine learning-based data science framework for evaluating the predictive validity of university entrance exams, focusing on first-semester academic performance across multiple cohorts. Using admission and academic records from 4,534 students at Jorge Basadre Grohmann National University (2017-2019), we constructed supervised learning models that control for instructor grading bias by engineering bias-resilient performance indicators. A broad set of classification and regression algorithms was benchmarked under 8-fold cross-validation, and feature importance analyses were applied to extract interpretable academic patterns. Results confirm that entrance exams, when evaluated through bias-controlled models, exhibit strong predictive capacity, with the overall entrance test score consistently emerging as the most influential feature. Subject-specific importance varied systematically by academic discipline, reinforcing the validity of track-specific cognitive profiles. This framework demonstrates how machine learning methods can enhance the fairness, validity, and analytical depth of entrance exam evaluations, offering a scalable and interpretable methodology for data-driven admission process improvement.

Instituciones: Universidad Nacional Jorge Basadre Grohmann