Tripoles: A New Class of Relationships in Time Series Data
Saurabh Agrawal (University of Minnesota);Gowtham Atluri (University of Cincinnati);Anuj Karpatne (University of Minnesota);William Haltom (University of Minnesota);Stefan Liess (University of Minnesota);Snigdhansu Chatterjee (University of Minnesota);Vipin Kumar (Univesity of Minnesota)
Abstract
Relationship mining in time series data is one of the research directions that is of immense interest to several disciplines. Traditionally relationships that are studied in spatio-temporal data are between pairs of distant locations or regions. In this work, we define a novel relationship pattern over three time series which we refer to as a \textit{tripole} that involves three time series. We show that tripoles can capture interesting relationships in the data that cannot be captured using traditionally studied pair-wise relationships. We propose a novel approach for finding tripoles in a given time-series dataset and demonstrate its computationally efficiency compared to the brute-force search on a real-world dataset from climate science domain. In addition, we show that tripoles could be found in real-world datasets from various domains including climate science and neuroscience. Furthermore, we found that most of the discovered tripoles are statistically significant and reproducible across multiple datasets that were completely independent to the original datasets that were used to find the tripoles. One of such discovered tripoles in climate data led to the discovery of a new climate teleconnection between Siberia and Pacific Ocean that was previously unknown in the climate domain.
