A load shedding scheme for frequent pattern mining in transactional data streams

In this paper, we study overload handling for frequent-pattern mining in online data streams. For a mining system with an e-deficient synopsis based algorithm, we propose a load shedding scheme to deal with the overload situation. The heavy workload of the mining algorithm lies mostly in the great deal of itemsets which need to be enumerated and counted by the mining algorithm. Therefore, our proposed scheme of load shedding involves the maintenance of a smaller set of itemsets, so the workload can be lessened accordingly. The unrecorded itemsets can be fast approximated for their counts when necessary. According to experimental results, the load shedding scheme can increase the throughput of the mining system and thus help manage the overload problem effectively to a certain extent.