Mining Frequent Patterns in the Recent Time Window over Data Streams

Because of the fluidity and continuity of a stream data, historic transactions might become obsolete and useless as new transactions arrive. It is more desirable to mine the frequent patterns in the recent time windows of a data stream. This paper proposes a method to mine the recent frequent patterns in the sliding window of data stream. It uses a conservative method to calculate the approximate frequencies of patterns in sliding window, and uses recent frequent pattern tree (RFP-tree for short) to incrementally capture the information of transactions in sliding window by scanning them only once. Moreover, based on the nice properties of an RFP-tree, a series of algorithms are designed to efficiently maintain and mine the frequent patterns from data streams. At last, the results of experiments show that the proposed method is more efficient and scalable than other existing algorithms.