2.1. Frequent pattern mining based on the sliding window model
Sliding window-based methods [24,25] have been proposed
to solve dynamic data stream problems of traditional frequent
pattern mining. Unlike traditional approaches that require two or
more database scans for mining patterns, the methods construct
their own data structures within a single database scan on
the basis of sliding window structures and perform pattern
mining operations. As one of the sliding window-based frequent
pattern mining methods, pWin [26] solved the problems of
previous algorithms consuming heavy memory by considering
characteristics of data stream environments where available
memory is limited and states of data are continually changed.
In addition, the algorithm used a prefix tree structure for
efficient tree searches. Another sliding window-based algorithm
using SWP-tree [20] employed a time decay model to discover
interesting patterns from historical transactions. Sliding windowbased
mining has been also utilized in other mining areas such
as maximal frequent pattern mining and top-k frequent pattern
mining. WMFP-SW [27] is an algorithm for mining maximal
frequent patterns over sliding window-based data streams. The
algorithm considers the latest data stream information and
weight conditions of items in the process of weighted maximal
frequent pattern mining. In addition, the sliding window model
can be applied in closed pattern mining [28]. However, the
above approaches only focus on the traditional pattern mining
framework but do not deal with erasable pattern mining.
Meanwhile, the proposed method can consider the erasable
pattern mining framework together with the weight conditions of
items and sliding window-based data streams.