Modelo predictivo usando Redes Neuronales de Boltzmann para el análisis del rendimiento a partir de los informes de cierre presentados por los docentes

Abstract

This undergraduate thesis presents NeuroCampus, a predictive analytics software based on Boltzmann Neural Networks to analyze teaching performance from closing reports and historical evaluation data from the Chemistry Department at Universidad Pedagógica Nacional. The proposal integrates principles of statistical physics, machine learning, and natural language processing into a reproducible architecture with a FastAPI backend and a React frontend, capable of transforming grades and comments into operational variables that support academic decision-making. The development included BM, RBM, and DBM models, together with a pipeline for data ingestion, validation, standardization, feature construction, sentiment analysis, and report generation. The results show that the system makes it possible to audit the complete cycle from raw data to aggregated indicators, improve process traceability, and provide useful estimates of expected performance, potential risks, and comparisons by lecturer, subject, and academic period.