Action rules mining. (Q1931698)

The book introduces several algorithms to learn so-called action rules, a form of rules proposed by Ras and Wieczorkowska, which describe how objects in a database could be re-classified when changing the value of some distinguished attribute(s). Starting from the notions of complete and incomplete information systems, the book shows how action rules can be either learned from classification rules or directly from the data sets in the database. The proposed algorithms can handle inconsistent data by applying concepts from rough set theory and they can also deal with incomplete attributes in the description of data sets. The learned action rules are further characterized by a degree of confidence expressed as a numeric value or attributes such as ``certain'' and ``possible'' as well as a cost function. By defining the cost and feasibility of an action rule, the author proposes a method to measure the ``level of re-classification freedom'' for objects in a decision system. Examples taken from medical application scenarios illustrate the power of the learning methods.