Study On The Association Rules Updating In The Data Mining

Data mining means a process of nontrivial extraction of implicit previously unknown and potentially useful information from data in database. Data mining tools can directly mining in large database for discovering potential knowledge in database . It plays an important role in decision support system, expert system and management information system. It may exploit all kinds of tools, such as, association rules, decision trees, artificial neural network, genetic algorithm, rule induction, and fuzzy logic .In above tools, association rules are mostly used. It is mainly introduced in the thesis that the definition, characteristic, classification, production and development of data mining .At the same time, it has finished the research of the typical algorithms of single level and multiple level association rules, the development of association rules updating and the basic research of association rules storing .The structure is: the technology of data mining ,association rules ,the research of typical algorithms of association rules ,the research of the algorithms of association rules updating ,the storage of association rules. It is studied in the dissertation that the contents of the following several respects emphatically. 1.The research of association rules: Since the association rules are proposed , numerous researchers have carried on extensive research of deepening to it, involve two respects mainly: On one hand, namely algorithm respect, including the proposition of new algorithm and improvement of the past algorithm . On the other hand, widening and extending to the concept and theorem of association rules. This thesis has introduced the concepts and theorems of association rules in detail. At the same time in order to improve the existing updating algorithm, it studied the typical single level of association rules algorithm Apriori in detail ; the multiple-level association rules algorithm ML-T2, and single level association rules incremental updating algorithm FUP, FUP * and multiple level association rules incremental updating algorithm MLUP. Method, realizing and performance of the existing typical algorithm mainly among them have been studied more in detail. 2.In the course of research the association rules updating, the mining of association rules often needs a large amount of repeated scanning work in a large-scale database, so the cost of finding the strong association rules is very large. But in practical application, the database is not static, it will be changing constantly with increase that the data are written down, and according to the needs of user , they often require the association rules to reflect the present state of the database , thus to the decision support, business administration they can offer the theoretical foundation. How the association rules should been effectively updated in the old database become newer problem in every perfect data mining system. what is called, it can find new knowledge in time not only can utilize existing knowledge but also to mean, but not operate the original algorithm of association rules in the database after being newer again .In this dissertation under the condition that consider that mini supports degree not to be changed, the association rules are updated high-efficiently when the trade records are added . It puts forward a kind of new incremental updating algorithm UA, this algorithm mainly tries every possible means to use priori knowledge, reduce the quantity of the new candidate sets, improve the efficiency of the algorithm at the same time. It is indicated through the repeated experiment that this algorithm has the advantages: the quantity of the new candidate sets is very small and the times of search database are few, so UA is a high-efficient algorithm with actual using value. 3. Obviously a large number of mined association rules are not very easy to deal with and store , certainly it is difficult that a person uses ,digests and understands. So, " the rule explodes ",a problem is becoming serious in the data mining system. This dissertation introduces some basic research work of the storing method of association rules following the international prevailing. In addition, in order to prevent the association rules from storing in the redundant rule while storing, it has been proposed: If a rule can be obtained from a simple rules set, then this rule is a redundant rule and needn't store . However, instead of relevant algorithms, the author is still in studying further. The research work of this thesis offers certain theoretical foundation and methods effectively to the existing association rules for further research and the association rules effectively being updated.