Association rules extraction and evaluation based on information theory

Large amounts of data are invading most sectors of activity (industrial, commercial, etc.), Meanwhile, methods for extracting knowledge from data have been developed in the last few years. It is often a question of finding models based on association rules in the following form: if A then B where A is the premise and B the conclusion. We must then select relevant rules by associating to each of them a value allowing the quantification of its relevance. Most data mining algorithms use the support and confidence of the rule as a measure of its quality. We propose similar indices, based on information theory, allowing the evaluation of these rules in terms of truth and efficiency of the rule. We then highlight theorems related to a set of generated rules in order to simplify them by eliminating redundancies.