Modelo predictivo para semaforización del rendimiento estudiantil

Abstract

Abstract: This thesis, "A Predictive Model for Student Performance Signals," focuses on the problem of dropout rates and poor academic performance in distance education at UTPL. Adopting the CRISP-DM methodology, this work addresses the objective of "developing and evaluating a specific predictive model that allows for the prediction of student performance, as well as the implementation of a signal-system mechanism that facilitates personalized and timely interventions." This work was conducted in the canton of Loja, with academic data from 32,956 students and 42 relevant variables. It uses advanced data analysis techniques, outlier processing, variable normalization, unsupervised learning (K-means clustering), and supervised learning (neural network) methods, which allowed students to be grouped based on their academic performance into three dropout levels (high, medium, and low). The overall conclusion of the research indicates that the developed model is effective in detecting patterns of academic risk. To this end, it offers an automated tool that contributes both to improving predictions of student academic performance and to facilitating intervention guidelines for UTPL students.