Virtual Choreographies of Learning: AI-Driven Discovery in K-12 Online American Schools

Abstract

Fully online K-12 education models, such as those provided by Connections Academy, serve over 100,000 students across the U.S., many of whom transition from traditional schools due to prior academic or personal challenges. While national reports often show lower standardized test scores for online learners, these environments enroll a significantly higher proportion of at-risk students, complicating fair comparisons. Despite the massive scale of online learning, there is limited granular research on how students actually navigate these platforms, how their study patterns evolve, and how these routines relate to their learning progress. This dissertation addresses these questions by applying the learning choreographies framework to detect recurring sequences of online learning actions. Due to the unavailability of proprietary institutional logs, the project adopted a data engineering approach, developing a simulator to generate a large-scale synthetic dataset designed to reflect the pedagogical rules and academic rhythms of a fully online K-12 environment. An end-to-end analytical pipeline was then constructed: unsupervised methods, primarily k-means clustering, were used to identify distinct behavioral profiles, while a supervised Random Forest model was trained to predict student success and extract early behavioral signals. The analysis surfaced four distinct learning choreographies, indicating that the quality and intent of engagement are associated with academic success far more than the sheer volume of clicks. The predictive model found that frequent, distributed study sessions and synchronous participation are strongly associated with passing, while it flagged "unproductive navigation", such as excessive dashboard refreshing and repeated grade-checking without content consumption as an early and critical risk signal. Ultimately, this work contributes a validated, data-agnostic analytical pipeline, fully prepared to process real institutional data and to support educators with actionable insights for early pedagogical interventions.