Good article. Even we use DT approach where ML scoring based decsion making seems over kill. But in real world data, decision tree doesn't plot in such clean way. Data classes are unbalanced eg. Only 1% percent purchased. It is difficult to extract simple rules. Any thoughts around that ?