Evaluating Algorithms for Predicting Academic Performance of Students with Disabilities

Abstract

This study analyzed the opportunities for predicting the academic performance of university students with disabilities through the use of machine learning algorithms. In the context of Information and Communication Technologies (ICT) as an essential tool for educational inclusion, the research aims to develop a reliable predictive model to optimize academic performance and reduce student dropout rates. The study is based on data from students at the National University of San Marcos (2019–2023), evaluating sociodemographic, academic, and contextual variables. Models such as Support Vector Machine (SVM), Logistic Regression, Decision Trees, among others, were implemented, using metrics such as precision, sensitivity, F1-Score, and AUC for evaluation. The findings highlight SVM as the most accurate model (AUC of 0.9839), followed by Logistic Regression, known for its balance between accuracy and interpretability. Although Random Forest and Neural Networks also showed good results, KNN presented significant limitations. The study concludes that predictive models are fundamental tools for identifying students at risk and developing personalized strategies that promote inclusion. Furthermore, it is recommended to explore advanced techniques such as deep learning and extend the implementation of these models to diverse educational settings.