Research On The Mining Frequent Itemsets Over Data Stream

In the end of the 20th century, with the advance of information science and technology, a new model named data stream appeared in many practical application fields, such as financial markets, the network monitor, and wireless sensor networks. Different from data sets in traditional static databases, data stream is continuous, strict and orderly in time dimensions and its value constantly changes. Due to the characteristic of data stream, traditional data mining algorithms, including frequent itemsets mining, are difficult to cope with data stream. So frequent itemsets mining over data stream is explored in depth. the main contents are as following:(1) the data stream model, the feature of data stream mining and the existing research results are summarized. Then, the technique of frequent itemsets mining over data stream is discussed, and a new model named transaction list group is proposed.(2) After the investigation of both traditional and data stream frequent itemsets mining, a new algorithm of frequent itemsets mining named DSTLG is proposed. The algorithm of DSTLG is based on the idea of the sliding window, the transaction list group and approximation to mine frequent itemsets.(3) In order to reduce the number of mining results, based on the idea of DSTLG, the expansion of DSTLG named DSMTLG is proposed to mine the maximal frequent itemsets over data stream.(4) Through a series of experiments, DSTLG and DSMTLG algorithm are proved to have an efficient time and space performance. In terms of time and space cost, relational analysis is also presented.