This study presents the preliminary validation of an integrated system that combines predictive analytics with an artificial intelligence (AI)-powered chatbot to address academic failure and dropout in higher education. The system leverages a high-performance Light Gradient Boosting Machine (LightGBM) predictive model, previously validated to achieve state-of-the-art accuracy (Area Under the Curve (AUC) = 0.953), to identify at-risk students at an early stage. The intervention component is an AI chatbot powered by the GPT-4o-mini large language model (LLM), which operates via WhatsApp® to deliver personalized, real-time support using context-aware semantic search. A pilot implementation with 108 students demonstrated the system’s viability, with 39 at-risk students identified and engaged in 742 interactions. Key operational metrics included a high student response rate of 89.7%, an initial intervention time of 6.6 seconds (s), and an average chatbot response time of 12.9 s. These findings confirm the technical feasibility of fusing machine learning with conversational AI to create timely, scalable, and personalized interventions, establishing a foundation for improving student success and retention in the future.
Building similarity graph...
Analyzing shared references across papers
Loading...
Felipe Emiliano Arévalo-Cordovilla
M. Pe
PeerJ Computer Science
Universitat Politècnica de Catalunya
Universidad Estatal de Milagro
Building similarity graph...
Analyzing shared references across papers
Loading...
Arévalo-Cordovilla et al. (Thu,) studied this question.
synapsesocial.com/papers/69a286600a974eb0d3c01502 — DOI: https://doi.org/10.7717/peerj-cs.3656