Research Of Fuzzy Association Rules Algorithm Based On Data-driven FCM

Cluster analysis is an essential approach in data mining while fuzzy association rule miningalgorithms is a major research direction in data mining. With the proposed fuzzy C-meansclustering algorithm (Fuzzy C-means, FCM),FCM algorithm that applied to mining fuzzyassociation rules become an important research area of data mining.This paper, mainly through in-depth study fuzzy C-means clustering algorithm and fuzzyassociation rules algorithm, first proposed the a pretreatment method FCM algorithm based ondata-driven, and then based on the traditional association rules mining Apriori algorithm andweighted fuzzy association rule algorithm proposed two new improved algorithm, the main work ofthis paper is as follows:1, FCM algorithm based on data-driven method of pretreatment (data-driven fuzzy c-means,DD-FCM) was proposed. This method can convert quantitative attributes to binary attributes,data-driven approach to generate fuzzy membership functions and fuzzy partitions. Theexperimental results show that the DD-FCM pretreatment method generated fuzzy membershipfunction can work properly in the case of no expert pre-given, and this method outperforms BIRCHand CLARANS clustering algorithms on pretreatment performance.Therefore, for large databases,the method is feasible and efficient.2, Apriori algorithm based on the DD-FCM (An Apriori Algorithm Baesd on data-driven fuzzyc-means, DD-FCMA) was proposed. Deal with the fuzzy sets are generated by the method ofDD-FCM, and use the property of Apriori algorithm to mining fuzzy association rules. As for largedatabases,it overcomes the shortcomings of the traditional Apriori algorithm in mining time. Underthis approach, the excavated rules has a strong correlation and semantic.3, An approach based on the DD-FCM weighted fuzzy association rule mining algorithm(Weighted fuzzy association rules based on data-driven fuzzy c-means, DD-FCMW) was proposed.The algorithm does not rely on expert given membership function, and solve the problem of theclosed down by the weighted association rules established. The experimental results show that thealgorithm is suitable for large databases contain Boolean and numeric data with good performanceand scalability.