記錄 49 之 6058
Record:   Prev Next
作者 Tzacheva, Angelina Assenova
書名 Action rules: Feasibility and lowest cost reclassification
國際標準書號 0496127586
book jacket
說明 125 p
附註 Source: Dissertation Abstracts International, Volume: 65-11, Section: B, page: 5843
Director: Zbigniew W. Ras
Thesis (Ph.D.)--The University of North Carolina at Charlotte, 2004
Two aspects of interestingness of rules have been studied in data mining literature: objective and subjective. Objective measures are data-driven and domain independent. Subjective measures, including unexpectedness, novelty and actionability, are user-driven and domain-dependent. A rule is actionable if user can do an action to his/her advantage based on the rule. This important definition is subjective and therefore rather vague, which presents a challenge in evaluating. In this study we claim that actionability can be seen as partially an objective concept. For that purpose, a new class of rules, called action rules, is used. These rules assume that attributes in a database are divided into two groups: stable and flexible. Flexible attributes provide a tool for making suggestions to a user to what changes can be made in order to re-classify objects to a desired state. In this study, the notion of support and confidence of action rules is introduced. Further, the notion of cost and feasibility of an action rule is proposed. A search graph based method for constructing feasible action rules at the lowest cost is given for a standalone information system. The method was applied in three different databases in two domains: medical and financial. Results show action rules of cost lower than the user is aware can be discovered, meeting the desired confidence threshold. We use the cost as an interestingness measure for action rules. Since cost in some cases can be calculated in objective manner, we claim that actionability can be seen as partially an objective concept. As a direction for the future, the application of the proposed algorithm in distributed autonomous information system is discussed
School code: 0694
DDC
Host Item Dissertation Abstracts International 65-11B
主題 Computer Science
0984
Alt Author The University of North Carolina at Charlotte
記錄 49 之 6058
Record:   Prev Next