Time Series Center   Harvard University

Outlier Detection

Description

Quasars, radio pulsars, and cosmic gamma-ray bursts were all discovered by alert scientists who, while examining data for a primary purpose, encountered aberrant phenomena whose further study led to these legendary discoveries. Such discoveries were possible when scientists had a close connection with their data. However, in the era of massive data sets, unexpected discoveries through manual inspection of data is improbable if not impossible. Fortunately, automated anomaly detection programs may be able to resurrect this mode of discovery and identify atypical phenomena indicative of novel astronomical objects. The detection of anomalous periodic light-curves (photometric time series data from the astrophysics domain) can focus attention on new and unusual items, with the potential to lead to the discovery of novel astronomical phenomena. Our data consist of N time series from multiple sources whose periods are not synchronized. Much of the prior research in time series anomaly detection assumes a long contiguous single-source time series, and is not applicable to our data. A challenge of our data is that a simple similarity calculation between any two time series results in a sub-optimal measurement unless a phase adjustment is performed. However, the cost of performing this adjustment for every pair of time series is expensive. Anomaly detection on astrophysics light-curve data necessitates a fast method that can handle the phasing issue described above, and the additional challenges of the data being voluminous, high-dimensional and noisy. We have developed PCAD, an unsupervised anomaly detection method that handles phase adjustments for large sets of periodic time-series data, and allows a flexible definition of anomaly. We report experiments on three classes of stars - Cepheid, Eclipsing Binary and RR Lyrae - as well as other periodic time series data sets. (See the paper Finding Anomalous Periodic Time Series: An Application to Catalogs of Periodic Variable Stars, below)

Publications:

• Variable Stars (Who Cares?) (TALK)
• Finding outlier light-curves in catalogs of periodic variable stars (PAPER)
• The IIC Time Series Center: How Astronomers, Computer Scientists and Statisticians together are working to tackle hard problems in Astronomy (SEMINAR)
• Introduction to Machine Learning Research on Time Series (LECTURE)
• Finding Anomalous Periodic Time Series: An Application to Catalogs of Periodic Variable Stars (PAPER)
• The IIC Time Series Center: How Astronomers, Computer Scientists and Statisticians are working together to tackle hard problems in astronomy (SEMINAR)
• Finding Anomalies in Periodic Time Series (POSTER)

Last Updated at: 2007-10-03 04:08 PM.