Assessment of the university emotional climate using convolutional models: A study in the Systems Engineering Degree Program – UPEA

Abstract

Background: The emotional climate in university settings is a key factor in academic performance, institutional coexistence, and psychological well-being. However, most studies are based on subjective instruments, such as surveys and self-reports, which limit the collection of objectives and real-time information. Objective: To evaluate the emotional climate of university students through the operational implementation of a pre-trained, proprietary convolutional neural network (CNN) model, applied in a face-to-face university setting within the Systems Engineering program at the Public University of El Alto (UPEA), providing empirical evidence on the contextualized and responsible integration of artificial intelligence in higher education. Methods: A quantitative, descriptive, non-experimental, cross-sectional study was conducted with multiple independent situational cuts (entry and exit), without individual follow-up. A proprietary CNN architecture was used, composed of three convolutional blocks with Batch Normalization and Dropout, accumulating approximately 1.7 million trainable parameters. The model was pre-trained with 35.887 grayscale facial images of 48×48 pixels (Data Set FER-2013), achieving a training accuracy of approximately 90% and an average validation accuracy of 65%. Evaluation metrics such as precision, recall, and F1 score were also used. In the field application, 1,850 emotional detections were recorded in the Systems Engineering program at the Public University of El Alto. Results: A predominance of neutral emotions (38.9%) was observed, followed by positive emotions such as happiness and surprise. Moderate variations were identified between arrival and departure, with a slight increase in emotions such as sadness and anger at the end of the academic day. Similar emotional patterns were also observed among the different institutional roles. Conclusions: The findings provide empirical evidence on the applicability of CNN models trained on standardized databases for the objective measurement of university emotional climate in real-world Latin American contexts, contributing to the field of artificial intelligence applied to higher education. Furthermore, the study proposes a replicable methodological approach for the automated monitoring of university emotional climate using artificial intelligence, opening new lines of research in educational analytics and institutional well-being