Hybrid time decay model and probability decay window model for data stream closed frequent pattern mining

Data stream is continuous and uncertain. Frequent pattern mining for data stream will cause that data distributes unevenly and concept drift. In order to improve mining efficiency and decrease data storage, we propose a hybrid time decay model and probability decay window model (HTPDWM) for data stream closed frequent pattern mining. This new method is divided into three steps. First, we adopt mining closed frequent pattern of sliding window model and time decay model in data stream to deal with new and old things. Second, we use probability decay window model and closure operator to calculate ex-pected support degree and improve efficiency of close pattern mining respectively. Third, we use decay factor to correct concept drift and data distributes evenly. Finally, we make experiments to verify the ef-fectiveness of the new method. Results show that HTPDWM can present stable with different sliding window and have better performance when processing time and memory space.