A sliding window algorithm for mining frequent itemsets on data stream

Sonic applications generate continuous, unbounded, and extremely fast data stream. Due to its characteristics, it is difficult to apply the existing productions of mining frequent itemsets in traditional databases to the data stream environment directly. A one-scan algorithm SFP of mining frequent itemsets on data stream was proposed by using a fixed-size sliding window model and adopting a data structure of SPFT. The window limited memory usage. SPFT made use of the advantages of FP-Growth and additionally introduced a pivotal timeldList into the SPFT. TimeldList marked the itemsets contained in those expired transactions so as to remove the effect of old transactions on the mining results as the window slides forward.