Discovering Patterns With Weak-Wildcard Gaps

Abstract

Time series analysis is an important data mining task in areas such as the stock market and petroleum industry. One interesting problem in knowledge discovery is the detection of previously unknown frequent patterns. With the existing types of patterns, some similar subsequences are overlooked or dissimilar ones are matched. In this paper, we define patterns with weak-wildcard gaps to represent subsequences with noise and shift, and design efficient algorithms to obtain frequent and strong patterns. First, we convert a numeric time series into a sequence according to the data fluctuation. Second, we define the pattern mining with weak-wildcard gaps problem, where a weak-wildcard matches any character in an alphabet subset. Third, we design an Apriori-like algorithm with an efficient pruning technique to obtain frequent and strong patterns. Experimental results show that our algorithm is efficient and can discover frequent and strong patterns.