The rate of college students getting D’s, F’s, or Withdrawing beyond a course drop deadline (known as the DFW rate) has been a national problem. The goal was to understand not only what drives DFW risk, but to create an early alert tool by predicting risk based on a combination of course performance and demographics data. Notable skills and tools included working with Canvas LMS’ APIs, feature engineering, Logistic Regression and Association Rules. It was also important to use demographic data to screen for and correct model bias.