Mining Frequent Patterns in Data Stream over Sliding Windows

Frequent pattern mining in data stream is an important task. Under the time decay model, this paper presents a new algorithm SWFP for mining frequent patterns over sliding windows. The new definitions of the infrequent, critical and frequent patterns which reflect the actual statistical property of each pattern within the sliding windows, grasp the real substance of mining process and help to improve the mining quality essentially. The support decay mechanism is designed not only to differentiate the current and history transaction, but also to make the online pattern maintain operation easily and accurately. The reasonable strategy for the pattern pruning periodically is used to make big cuts in the maintenance cost and the error controlled in a small bound. Theoretical analysis guarantees no false negatives of SWFP. Experimental evaluation over a number of synthetic data sets demonstrates the efficiency and scalability of our method.