An Adaptive Frequent Itemset Mining Algorithm for Data Stream with Concept Drifts

Mining frequent itemsets in data streams has became one of the hottest research topics in data mining nowadays, recent algorithms that make use of definite error bound or probabilistic error bound, have relieved the temporal-spatial complexity at some extent. However, the introduction of unwanted sub-frequent itemsets, and the changes of itemsetspsila supports, namely concept drifts, lower the efficiency and the accuracy. In this paper, an adaptive frequent itemset mining algorithm for data stream with concept drifts is proposed. By monitoring the change of support, it measures the stabilities of supports, thereby adaptively adjusts the sampling periods. With biggish probability, the error of support could be upper bounded. The theoretical analysis and experiments prove its efficiency and accuracy.