The Application Of Fuzzy Forecast Model In Dynamic Association Rule Mining

With the rapid development of computer, communication, network technology and the wide use of automation technology in daily life, multi-type of data is being produced and collected with the unprecedented speed. Data mining reveals unknown implicit information and validates known law via analyzing data stored in the database, which assists the decision making. In the field of data mining, association rules, one of the earliest and most active research direction, has become a current research hot spot.The early association rule mining, a kind of static mining method that based on the transaction database, ignores the time factor, and regards rules that excavated will be permanent. But after continuous application and research, we summarize the results that rules excavated based on database in real life has close correlation with time. Therefore, in order to further describe characteristics of the rule changes over time, researchers have proposed the concept of dynamic association rule mining. Dynamic association rules demarcate transaction data sets based on year, month, date and other time granularity, and join the support vector and confidence vector as a new evaluation index of the rules. However, while taking of the time factor, the traditional rules mining based on existing data cannot ensure the timeliness of the rules in the future. In addition, many researchers will study the dynamic association rule mining focused mainly on the improvement of mining algorithm. Recently, it rarely comes to the reliability issues of rules that excavated. Therefore, in order to improve the quality of excavation, the dynamic association rule mining on further research is necessary.Based on the research of fuzzy sets and related theory, this paper combined the advantages of fuzzy forecast of high prediction accuracy and small sample modeling with dynamic association rule mining. Firstly, aiming at the existing problem of fuzzy time series method, an algorithm of fuzzy time series forecasting model based on clustering and partitioning is proposed. And through the experiments which compared with the other methods, this paper proves the effectiveness of the algorithm, and then on the basis of combining the theory of Markov chain, two kinds of combination forecast model named Fuzzy-Markov and Fuzzy–Gray are established. Through the modeling analysis and experiment comparison of different types of data sets, it is proved that in support count forecast of small sample size, Fuzzy-Grey combination prediction model for comprehensive the knowledge of each single model and showed a stronger practicability and higher prediction precision; Finally based on online shopping customer behavior analysis of actual case studies and combined with the Fuzzy-Grey prediction model, a dynamic association rule mining model based on data set of trading peak hours is established, and eventually dig out the rules that has guiding significance.Through the analysis of the mining rules meaning and the prediction of support variation trend, we find that the model established in this paper has good prediction effect for the support sequence of dynamic association rule and can dig up some more potential and useful rules compared to conventional mining methods. This shows that applied the fuzzy prediction model in dynamic association rule mining can be further in-depth analysis of rule changes and grasp the specific trends of rules over time, so as to achieve the purpose of further improve the quality of dynamic association rule mining.