This project explores student behavioral, textual, and limited demographic data retrieved from Michigan Virtual School for the 2014-2015 and 2015-2016 academic years. The primary method of analysis was deep learning (DL) however a variety of other data mining techniques were explored, including text analysis, to improve prediction accuracy. DL was also compared to machine learning (ML), and results indicate that DL was slightly better than other ML models; also the inclusion of textual content improved the overall predictive accuracy in identifying at-risk students. Factors affecting the predictive power of the analyses are discussed as well as recommendations and considerations for using this and similar predictive models in practice to identify at-risk students.