Mining fuzzy weighted association rule with domain knowledge

Abstract

The objective of data mining is to discover and extract valuable information from data, i.e., the hidden gold. Before we design a data mining system to support decision-making, we have to know how people make decision. Some sociologists carried out investigations into 60 highly successful California enterprise managers. These managers were asked how they made business decision. All of them answered that they used a combination of their intuition and the data. No one said they made decision based on data alone. Significant data mining problems have been addressed recently, namely, mining fuzzy association rules. Using the fuzzy set concept, the discovered rules are more easily understandable. In most models of mining fuzzy association rules, items are considered to have equal importance. However, depending on a user's domain knowledge, items should have different degree of importance for each user. Existing models do not address such individual interest and preference. To improve such models, we proposed the mining of fuzzy association rules with weighted items. The weighted support and weighted confidence for fuzzy association rules are defined. Our Experiments are based on some benchmark problems. For example, adult records selected by a census of the United States in 1990, show the strengths of our proposed methodology. The strengths are that, for different users who are interested in different attributes (i.e. who assign different values of weights to corresponding attributes), different fuzzy association rules which reflect the initial interest of users are mined. The relationship between the mining with and without weighted-items in a large database is obtained.