Discovering e-action rules from incomplete information systems

Abstract

s Abstract— Action rules, interventions, and E-action rules are examples of knowledge discovery tools for reclassification of objects and for defining actionability as a partially objective concept. However, most of these tools can only deal with incom- pleteness represented as null values. The purpose of knowledge discovery systems is to extract knowledge which is interesting and often interestingness is linked with actionability. In this paper, we present a new algorithm, DEAR 3, to discover actionability knowledge from an incomplete information system. 1 I. INTRODUCTION E-action rules (5), (6), are a key tool used for extracting higher level actionable information from large volumes of data. It can be applied in many real-life applications as a powerful solution to a reclassification problem. The basic principle of reclassification is a process of learning a function that maps a class of objects into another class by changing values of some of the classification attributes describing that class. The classification attributes are divided into stable and flexible. Saying another words, reclassification is a process of showing what changes in values in some of the flexible attributes for a given class of objects are needed in order to shift them from one decision class to another more desired one. E-action rule is a rule of the form ((ω) ∧ (α → β)) ⇒ (φ → ψ), where ω,(α → β), and (φ → ψ) are events. It states that when the fixed condition ω is satisfied and the changeable behavior (α → β) occurs in a database tuples so does the expectation(φ → ψ). Support and confidence are used to measure the importance of each rule to avoid generating irrelevant, spurious, and insignificant rules. Any E-action rule forms workable strategy that can be used in a business decision making process to increase profit, reduce cost, etc. Each E-action rule can be constructed by comparing pairs of previously discovered classification rules from a given decision system. The concept of E-action rule was introduced in (9) to enhance action rules (4) and extended action rules (5),(7), and (8) to extract actionable knowledge from databases containing nominal data. Several efficient algorithms for min- ing E-action rules have been developed (7), (8), (6), and (9). In all these papers, mining for action rules from a complete data is well understood and investigated on both the algorithmic and conceptual level. In this paper, we present a new algorithm DEAR 3f or discovering action rules. It consists of several basic steps: process of discovering classification rules, analyzing them, and process of action rules construction. The development of workable strategies for implementing them is naturally linked with the last step. We also propose a novel method CID for generating classification rules from an incomplete information system. The definition of an incomplete information system that allows to have non-singleton subsets as values of attributes is given. After forming classification rules, there are two possible options to construct E-action rules: one based on action forest (9) and the other one based on action tree (6). They both speed up the process of E-action rule construction. They differ in the splitting criterion applied at every node of a generated tree. In this paper, action-tree algorithm is utilized. We focus on mining E-action rules, since they are more meaningful, simpler, easier to interpret, understand, and to apply than classical action rules (9).