From data to classification rules and actions

Abstract

Action rules (or actionable patterns) describe possible transitions of objects from one state to another with respect to a distinguished attribute. Strategies for discovering them can be divided into two types: rule based and object based. Rule-based actionable patterns are built on the foundations of preexisting rules. This approach consists of two main steps: (1) a standard learning method is used to detect interesting patterns in the form of classification rules, association rules, or clusters; (2) the second step is to use an automatic or semiautomatic strategy to inspect such results and derive possible action strategies. These strategies provide an insight of how values of some attributes need to be changed so the desirable objects can be shifted to a desirable group. Object-based approach assumes that actionable patterns are extracted directly from a database. System DEAR, presented in this paper, is an example of a rule-based approach. System ARD and system for association rules mining are examples of an object-based approach. Music Information Retrieval (MIR) is taken as an application domain. We show how to manipulate the music score using action rules. © 2011 Wiley Periodicals, Inc.