Research On The Problem Of Association Rule Mining In Incomplete Relational Database

Data Mining is a field of increasing interest combining database, artificial intelligence and machine learning. The purpose of data mining is to facilitate understanding large amounts of data by intelligently and automatically discovering useful information and knowledge. Discovery of Association Rules is an interesting subfield of data mining. Historically, the problem of association rule mining was first introduced by Agrawal etc.in 1993, the motivation for it came from the desire to analyze large amounts of supermarket basket data(et customer transaction database ) and find some regularities for business decisions. The information maintained for the different transactions is the sets of items bought by each customer. In this case, it may be desirable to find how the purchase behavior of one item affects the purchase behavior of another. Association rules help in find such relationships accurately. Many researchers have shown great interest in the problem. They have made deeply research in the area. Meanwhile, the problem of association rule mining has been extended and applied to many other field, such as telecommunication, finance etc, which all have achieved good effect. Missing value have not been considered of interest since the problem of association rule mining was first introduced because association rules have been first developed to explore transaction database where the problem of missing values does not practically exist. However this problem becomes important if we try to find associations between values of different attributes in relational database where missing values are often inevitable. It is not obvious how to compute association rules from such incomplete database. In this paper, several typical algorithms of bool association rule mining, MS algorithm, SETM algorithm, Apriori algorithm and DIC algorithm, are introduced in detail and compared; the classical flI ~ ~t7~ THESIS method and key techniques including the discretion of quantitative attribute, the interestingness of association rules, and the processing of counting support of candidates during scanning database, are also introduced in detail; a new method of mining association rule in relational database based on equivalence class in rough set theory is presented, in which the problem of association rule mining, the support and confidence of the association rule are redefined; by investigating properties of incomplete relational database, the association rule induced from an incomplete relational database and how to estimate support and confidence of it are introduced; Meanwhile, the definitions of expected support and confidence of an association rule is presented, which guarantee some required properties of association rules mining; several previous solutions to the problem of missing attribute values have been discussed, by analyzing the shortcoming of these solutions and combining the defines above , a new define and method of association rule mining in incomplete database are presented finally.