Mining emerging patterns from time series data with time gap constraint

Discovery of powerful contrasts between datasets is an important issue in data mining. To address this, the concept of emerging patterns (EPs) has thus been in- troduced by Dong and Li. EPs are a set of itemsets whose support changes significantly from one dataset to another. Although an increasing number of works focus on this topic with regard to relational databases, few have considered mining EPs in time series. In this paper, we thus propose a framework named PIPs-SAX for mining EPs from time series data. The framework contains two phases: the first phase is data transformation and the second is the EPs mining. The first phase transforms the time series data into a symbolic representation based on the SAX and PIPs algorithms. In the second phase, we propose an algorithm, called TSEPsMiner, to mine time series EPs with a time gap constraint. Experiments on financial data collected from the Taiwanese stock exchange were also made in order to evaluate the effectiveness of the proposed framework.