FCAD: Feature-based Clipped Representation for Time Series Anomaly Detection

摘要: Since time series are a ubiquitous type of data, more and researchers investing in data mining recent years. It is generally known that have the characteristics large volume, high dimensionality ever-increasing. Thus, reduction usually first step mining. Over years, amount high-level representation approaches been proposed. In this paper, we propose novel bit level approximation called Feature-based Clipped Representation (FCR), similarity measure for FCR which lower bounds Euclidean distance introduced. Finally, an anomaly detection approach series, FCAD, based on corresponding measure. Extensive experiments conducted to demonstrate advantages FCAD detection.