The Research On Association Rule & Its Application In Data Mining

This paper concentrates on researching association rules mining theory and its algorithm model applied in information tables based on rough set theory. Firstly it gives a detailed introduction to the definition, technologies and study orientations of data mining. Association rules mining is one of the most study branches and its mining methods are discussed. Aiming at large amounts of rules in general methods, this paper places an emphasis on quality measures about association rules, and proposes an improved algorithm based on profit constraint. Further more, a method called MDRBapriori algorithm is produced to obtain default rules with the specified support and confidence thresholds in decision tables, and in this algorithm association rules model is used for reference. The main research of this paper is as follows:In the first place, this paper gives the definition and significance of association rules mining and investigates its mining process. The existing problems and its directions are also discussed here.In the second place, all subject and object methods about rule quality measures have been proposed by scholars to deal with some problems in association rules mining algorithms. In this paper, the industrial information — profit concerned by an enterprise — is introduced to be a constraint so that the redundant rules can be further eliminated and those more interesting rules can be shown users. Additionally, the parameter like profit extends the containing power of association rules, which will not simply reflect the trend between commodities, and from which the decision-maker can acquire more information.The model of support-confidence is adopted in all existing algorithms, which emphasizes on large item sets and usually ignores the small ones. However, to mine all association rules meeting the profit constraint ones expected by users, the rare data neglected previously will probably be interesting because of industrial information. A new algorithm is proposed, in which rare data are can be mined with relative support and then profit will be a constraint in the mining process. So the algorithm can mine all such association patterns meeting the requirement of profit whether their supports are high or not.Thirdly, the basic concepts about Rough set theory and reduction methods for attributes are given to a brief introduction. An approach called projection algorithm, is proposed by Mollestad and A. Skowron to obtain default rules. Its framework and performance is analyzed in detail.Subsequently, aiming at some defects of projection algorithm, this paper presents a new approach, called MDRBapriori algorithm, in which default rules extended can be abstracted in decision tables.After analyzing both projection algorithm and association rules algorithm models, we find they have the same essence of seeking solutions;that is to say, they seek all the possible combination patterns for all the attribute values. As a result, the solutions produced by both methods are approximately identical with respect to the same constraints. In terms of significance in statistics, this algorithm is proved that it can be replaced by certain association rule mining methods. Nevertheless a new approach called MDRBapriori, based on Apriori algorithm is proposed to generate default rules in the place of the projection algorithm, and has some particular advantages. This method does not need reduction, and it can acquire complete rules for the decision-maker to make a reasonable choice. The validity of the proposed algorithm model is proved by the experimental data. In addition, the operation of this algorithm model is convenient because of the maturity of association rules mining techniques. Some popular data mining tools such as DBMiner2.0 can be used to get default rules by the corresponding function on its association rules mining.Finally, a recapitulative conclusion is given, and the future research directions are proposed.