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

Abstract

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