Research On Dynamic Association Rule Mining Based On Grey System Theory

Data mining is a process of revealing the implied known unknown or validation rules byanalyzing data in large database or data warehouse. It can be roughly divided into manymodels such as classification, regression, association rules, time series and cluster amongwhich association rule model is the most widely used method and it’s a hotspot of research inrecent years at the mean time.The traditional association rule mining is a kind of static mining based on transactiondatabase. But in the process of mining it’s not fully considering the characteristics of thechange over time. After a long period of practical application and study, it’s known that theassociation rules mined in the actual database are often with time characteristics. So it’snecessary to join the time characteristic to the mining process to observe the changes of rulesover time.The dynamic association rule mining was born considering that the rules change overtime. It sets boundaries on data sets to describe the time characteristic of the rules. Butbecause of that the traditional dynamic association rule mining methods are based on existingdata, it’s difficult to know whether the strong dynamic association rules are still valid in thefuture. And at the same time, the current dynamic association rule mining researches aremainly focus on improving algorithm and less involved in the quality of mining in temporaldatabase. So it’s necessary to do deepen research on dynamic association rule mining toobtain information with high quality and real value.This paper proposed a dynamic meta-rule association rule mining algorithm based onGrey-cycle extension model and a method of tendency mining based on Grey Markov modelin dynamic association rule by combining the grey system theory, the dynamic associationmeta-rule mining and tendency mining together on the basis of the relevant knowledge. Bysetting up different models in view of different types of data, the given two algorithms fullytake advantages of grey model when the original data is small and have a better effect for datamodeling and prediction. And by applying the proposed methods in actual mining, they areproved to be valid and have higher prediction.By comparing the proposed methods with the traditional dynamic association meta-rulemining methods and tendency mining algorithm, the two methods are proved can min higherquality and stronger rules. And to a certain degree they overcome the blindness of mining andcan maximum min effective and potential rules to effectively improve the efficiency of thedynamic association rule mining and reference value of the results. And different with thetraditional algorithm, the proposed methods can get the rules with the change over time byanalyzing the tendency of the rule through the mathematical models.