Primitives in time series mining: Algorithms and applications
Abdullah Mueen · The Journal of the Acoustical Society of America · 2016
Time series patterns are waveforms with properties useful for various data mining tasks such as summarization, classification and anomaly detection. In this talk, I present three types of time series patterns: Motifs, Shapelets, and Discords. Motifs are repeating patterns that repeat in seemingly random time series data; Shapelets are small segments of long time series characterizing their sources; Discords are anomalous waveforms in long time series that do not repeat anywhere else. I briefly discuss efficient algorithms to discover these patterns and present cases in mining data from robots, humans and social media. Cases include activity classification using accelerometer data, correlated clusters in social media data, and anomaly in astronomical data.