Predicción del rendimiento académico como indicador de éxito/fracaso de los estudiantes de ingeniería, mediante aprendizaje automático

Abstract

This research study identifies variables that influence the prediction of performance in industrial engineering undergraduate students at the Universidad Distrital (Colombia) by three methodologies: filter, wrappers, and integrated. Python programming language classification algorithms such as decision tree, K nearest neighbors, and perceptron are implemented and they are compared to obtain the best prediction results. The results show that gender and the ICFES Score (Colombian nation-wide university admission exam) for mathematics were in the upper range in all the selection methods. The Perceptron algorithm is the most accurate of all the algorithms tested. It is concluded that the variables that most affect academic performance in engineering students are: age, gender, tuition fee, the overall ICFES score, and the ICFES scores for mathematical aptitude and cohort mathematics.