| Data models built for analyzing student data often obfuscate temporal relationships for reasons of simplicity, or to aid in generalization. We present a sequence model that is based on temporal relationships in heterogeneous student data as the basis for building predictive models to identify and understand students at risk. The properties of our sequence data model include temporal structure, segmentation, contextualization, and storytelling. To demonstrate the benefits of these properties, we have collected and analyzed 10 years of student data from the College of Computing at UNC Charlotte in a between-semester sequence model, and used data in an introductory course in computer science to build a within-semester sequence model. Our results for the two sequence models show that analytics based on the sequence data model can achieve higher predictive accuracy than non-temporal models with the same data. The sequence model not only outperforms non-temporal models to predict at risk students, but also provides interpretability by contextualizing the analytics with the context features in the data model. This ability to interpret and explore the analytics, enables the development of an interactive exploratory learning analytics framework to involve the domain experts in the process of knowledge discovery. To show this potential of the sequence model, we developed a dashboard prototype and evaluated the prototype during focus group with our college faculty, advisors, and leadership. As a result, the dashboard facilitates generating new hypotheses about student data, and enables the discovery of actionable knowledge for domain experts. |
