An adaptive approximation method to discover frequent itemsets over sliding-window-based data streams

Frequent-pattern discovery in data streams is more challenging than that in traditional databases since several requirements need to be additionally satisfied. For the sliding-window model of data streams, transactions both enter into and leave from the window at each sliding. In this paper, we propose an approximation method for mining frequent itemsets over the sliding window of a data stream. The proposed method could approximate itemsets’ counts from the counts of their subsets instead of scanning the transactions for them. By noticing the more dynamic feature of sliding-window model, we have made an effort to devise a promising technique which enables the proposed method to approximate for itemsets adaptively. In addition, another technique which may adjust and correct the approximations is also designed. Empirical results have shown that the performance of proposed method is quite efficient and stable; moreover, the mining result from adaptive approximation (and approximation adjustment) achieves high accuracy.