Comprehensible Knowledge Discovery: Gaining Insight from Data
Michael J. Pazzani · 1997
this paper, we argue that existing data mining systems fail to realize the full potential benefits of data mining because they do not seriously address the issue of the comprehensibility of learned models. We argue that models that are minor variants of what is already known are preferable to models that differ drastically from current understanding when the predictive power of such models are equivalent. We describe extensions to a learning system, FOCL and demonstrate that these extensions produce learned models that are more useful to experts. To illustrate our point, consider the following simple examples of economic sanctions incidents: