Decision Trees Predictive models fall into one of two general categories: regression models that predict continuous values and classifi- cation models that predict discrete values. Our first experiments focused on regression models, but we found that they performed poorly since coverage is bounded by [0%, 100%]. Many closely-related nominal metrics or an unusually large or small numeric metric could send the prediction out of this range, causing a nonsensical prediction less than 0% or greater than 100%. We then experimented with several classification models and found that decision trees best served our purposes for the following reasons: