Analysis of machine learning algorithms to evaluate the academic performance of students in the subject of Mathematics

Abstract

The academic performance of students is crucial for the educational and social development of a country, as it directly impacts the future opportunities of young people and the collective progress of society. Mathematics plays a fundamental role in basic education, yet many students face difficulties that affect their academic performance and can limit their long-term development. This challenge is reflected in the results of the Program for International Student Assessment (PISA), which highlights the gaps in mathematics learning globally. The present research seeks to identify the most accurate machine learning algorithm for predicting the academic performance of high school students in mathematics. The approach adopted is quantitative, with a non-experimental, applied-type design. The sample included 400 students, whose data were obtained from a public repository at the University of California, Irvine. The data collected covered demographic, academic, psychological and social factors, among others, totaling 30 key variables. These data were used to train and evaluate various machine learning algorithms to identify the most important determinants in the learning process. The most accurate algorithm was the Dense Neural Network, reaching 74% accuracy, demonstrating its effectiveness in predicting academic performance. The results revealed a positive and significant relationship between the variables analyzed and academic performance, suggesting that improvements in these variables could increase the students' academic average.