We created holistic profiles of student interaction by designing: time series analyses to capture the timing of when a student interacted, NLP analytics for unstructured text to quantify the complexity of what was said, network analyses to detail a student’s involvement within a class, and statistic based heuristics to normalize performance compared to their peers. We were then able to identify clusters of students with equivalent performance, shared behaviors, and similar learning styles. Ultimately these profiles and clusters were used to train a decision tree based model to predict which students are most at risk of poor future performance and why, providing teachers with information they could use to improve student retention.