We have been running trials on a 16 question survey, predicting the responses to a particular question using other data in the survey. What we discovered is that the more rules we allowed KL analytics engine to produce, the lower the accuracy and the harder it becomes to explain the model to another person.
