Discovering Fuzzy Partial Periodic Patterns in Quantitative Irregular Multiple Time Series

Partial periodic patterns are an important class of regularities in multiple time series data. Most previous works focused on finding these patterns in a binary series by disregarding the quantities of objects. This paper explores the concept of “fuzzy sets” and proposes a novel model of fuzzy partial periodic patterns (F3Ps) that may exist in a quantitative series. F3Ps have value because they represent regularities that are predictable in a series. Unfortunately, finding F3Ps is challenging due to its colossal search space. We introduce a novel pruning technique to reduce the search space and computational cost of finding these patterns. We also present an efficient depth-first search algorithm, F3P-Miner, to find all F3Ps in a series. We also describe a new pattern representation technique, temporal ordering and grouping with a wildcard character, to visualize long patterns easily. Experimental results demonstrate that the proposed algorithm is efficient. Finally, we present the real-world applicability of F3Ps by finding helpful information about polluted areas in the one year Japan's air pollution database.