Predicción de la deserción estudiantil Universitaria utilizando Deep Learning. Caso: Instituto Superior Tecnológico Quito

Abstract

Abstract: In Ecuador, higher education institutions face a significant challenge: high student dropout rates. This issue negatively impacts both the institutions and the students along with their families. Transformer models have recently stood out due to their effectiveness. Although initially designed for natural language processing, Transformers are increasingly utilized in various fields such as climate predictions, financial forecasting, and image and video recognition. The main objective of this study was to design a Transformer model capable of predicting student dropout at Instituto Tecnológico Quito. Historical student data were analyzed to detect patterns that allow the model to identify students at risk of dropping out in the future. Using the LIME technique, several key factors influencing dropout were identified, including gender, marital status, economic condition, academic program, class schedule, employment status, and career changes. It is crucial to highlight that dropout prediction does not depend solely on a single variable but rather on the combination of all these characteristics. Upon evaluation, the model achieved 96% accuracy, making it an effective predictive tool.