RF03-08 SAAT 2.0: A Vectorial and Unsupervised Learning Framework for Academic Analytics and Early Warning Systems Details

Start Date: 2026-06-08

End Date: 2026-06-08

Early Warning Systems (EWS) in higher education traditionally rely on predefined indicators and supervised predictive models. Although effective in specific scenarios, these approaches often fail to discover emerging behavioral patterns and hidden academic trajectories. This paper proposes SAAT 2.0 (Academic Analytics and Early Warning System 2.0), a mathematical framework that combines hierarchical vector representations, temporal analytics, unsupervised learning, similarity analysis, and academic trajectory mining. The model extends traditional academic analytics by introducing two additional spaces: an Academic Embedding Space and a Pattern Discovery Space. Through these extensions, students are represented not only by observed variables, indicators, and risks, but also by latent behavioral structures discovered directly from institutional data. The proposed framework enables clustering, similarity search, trajectory analysis, emerging risk detection, and explainable academic intelligence. Keywords—Learning Analytics, Early Warning Systems, Academic Embeddings, Unsupervised Learning, Educational Data Mining, Student Trajectories.