A machine learning model to predict standardized tests in engineering programs in Colombia - Predicting test results for student reinforcement decisions

Aug 3, 2026

Research develops a machine learning model to forecast results of Colombia’s national standardized tests for engineering programs, enabling student-level predictions that support reinforcement strategies to improve performance. A Learning Analytics pipeline is organized into three stages: database analysis and debugging, multivariate analysis, and machine learning techniques. Results reveal an association between high school performance levels and university outcomes, with the Generalized Linear Network Model showing the best fit. Model training and evaluation metrics are reported using Accuracy, AUC, Sensitivity, and Specificity.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages8

Reading Time20 min

AccessFree to download

Introduction

Educational quality and accreditation context

Role of standardized tests in Colombia

Learning analytics approach

Database analysis and debugging

Multivariate analysis

Machine learning modeling

This article has been accepted for publication in IEEE Revista Iberoamericana de Technologias del Aprendizaje. This is the author's version which has not been fully edited and content may change prior to final publication. Citation information: DOI 10. 1109/RITA.2023.3301396 OASP-RITA-09-2021-0077.R1 A machine learning model to predict standardized tests in engineering programs in Colombia Misorly Soto-Acevedo, Alfredo M. Abuchar-Curi, Rohemi A. Zuluaga-Ortiz, Enrique J. Delahoz-Domínguez* Forecasting of Standardized Test Results for engineering students through Machine Learning Abstract— This research develops a model to predict the results of Colombia's national standardized test for Engineering programs. The research made it possible to forecast each student's results and thus make decisions on reinforcement strategies to improve student performance. Therefore, a Learning Analytics approach based on three stages was developed: first, analysis and debugging of the database; second, multivariate analysis; and third, machine learning techniques. The results show an association between the performance levels in the Highschool test and the university test results. In addition, the machine learning algorithm that adequately fits the research problem is the Generalized Linear Network Model. For the training stage, the results of the model in Accuracy, AUC, Sensitivity, and Specificity were 0.810, 0.820, 0.813, and 0.827, respectively; in the evaluation stage, the results of the model in Accuracy, AUC, Sensitivity, and Specificity were 0.820, 0.820, 0.827 and 0.813 respectively. Index Terms— learning Analytics, Machine Learning, Predictive Evaluation, standardized tests. 1. INTRODUCTION Quality, when viewed as a process, can be objectively gauged through performance indicators. For instance, longitudinal analysis of standardized tests [1], or the interplay between economic variables, infrastructure, and academic outcomes [2] can offer valuable insights. Hence, to achieve educational quality, implementing ongoing selfassessment policies is necessary, aiming towards a continuous improvement process [3] . Thus, quality should be objectively evaluated. Internationally, one method of estimating quality in education is through the Accreditation Board of Engineering and Technology (ABET), a nongovernmental, non-profit entity comprising technical and technological societies. These societies establish the policies of the process and accredit programs in applied sciences, computing, engineering, and engineering technologies both within and outside the United States [4] . Currently, Manuscrito recibido el día de mes de año; revisado día de mes de año; aceptado día de mes de año. English versión received Month, day-th, year. Revised Month, day-th, year. Accepted Month, day-th, year. Misorly Soto-Acevedo, Facultad de Ciencia Básicas, Universidad Tecnológica de Bolívar, Cartagena, Colombia ([e-mail: ](e-mail: msoto@utb.edu.co)[msoto@utb.edu.co](e-mail: msoto@utb.edu.co)) Rohemi A. Zuluaga-Ortiz, Facultad Ingeniería, Universidad del Sinú, Cartagena, Colombia (e-mail: [rohemi.zuluaga@unisinu.edu.co](rohemi.zuluaga@unisinu.edu.co)) Alfredo M. Abuchar-Curi, Facultad de Ingeniería, Universidad Tecnológica de Bolívar, Cartagena, Colombia (e-mail: [aabuchar@utb.edu.co](aabuchar@utb.edu.co)) *Enrique J. Delahoz-Dominguez, Department of Productivity and Innovation, Universidad de la Costa, Barranquilla, Colombia (e-mail: [edelahoz13@cuc.edu.co](edelahoz13@cuc.edu.co)) numerous universities are undergoing not only national accreditation processes but also international ones, which requires measurements at different stages of the learning process and compliance with the demanded quality standards. Standardized tests are the principal means used in Colombia and globally to measure academic achievement [5] . In Colombia, standardized tests measure the academic achievement of students in secondary education (Saber 11) and

FAQs

What is the document’s main objective?

To build a machine learning model that predicts results of Colombia’s standardized tests for engineering programs, supporting reinforcement decisions to improve student performance.

How is the Learning Analytics approach structured?

It follows three stages: database analysis and debugging, multivariate analysis, and machine learning techniques.

Which model is reported as fitting the problem best, and what evidence is given?

The Generalized Linear Network Model is highlighted, with reported Accuracy, AUC, Sensitivity, and Specificity on both training and evaluation stages.

Aug 3, 2026

Upload by Emma Wilson