Abstract
The Discovery of production rules (association rules and/or classification rules) is one of the most important tasks of data mining. The discovered knowledge is intelligible and comprehensible by experts in any field. In previous works, we used formal concept analysis to discover classification rules [4] and association rules [5] embedded in data sets.
One of the difficulties we found is to measure the pertinence of the discovered rules. In supervised learning of classification rules, we used the known entropy measure [3]. In un-supervised learning of association rules, we used the known support measure [2]. However, some recent works [11, 9] have proven the insufficiency of these measures and have introduced other ones.
In this paper, we present a bibliographic summary of many existing pertinence measures. Then, we present an experimental study of the behavior of these measures in order to help the users of our learning system, choosing the appropriate measure.