Research On Algorithm For Mining Maximal Frequent Itemsets Over Data Streams

Data streams mining is a hot topic of applied mathematics and database. One of the core issues in data stream mining is frequent itemsets mining. Since maximal frequent itemsets contain frequent itemsets represented by their subsets, the storage space can be reduced to the greatest extent. So algorithms of mining maximal frequent itemsets over data streams is studied in this paper, main work is as follows:(1) Some algorithms of data streams mining are studied and improved, problems are pointed out. So during the process of counting itemsets in DSM-MFI, decay rate is imported to get an algorithm of mining maximal frequent itemsets over data streams called DSM-AMFI. This algorithm reduces the impacts of timed info to current mining results by reducing the counting of historical data. So the mining scale is reduced and more dynamic algorithm is achieved. DSM-AMFI shows better time efficiency under the sensitive data stream.(2) Structure of prefix binary tree-PBT is designed, combined with structure of Bitset, way of classic storage is changed. Furthermore, an algorithm is designed which is called DSMMFI-BPBT based on prefix binary tree. During the process of mining maximal frequent itemsets, since bitwise operation is done only to nodes of PBT, bit string created during the process of bitwise operation is not necessarily kept in memory. So times of bitwise operation are reduced, and storage size is reduced as well. The experimental results indicate that the algorithm in the paper shows better efficiency both in time and space. Better time efficiency compared with DSM-MFI in the occasions of long average length of mining and low support threshold is achieved...