Survey on Outlier Pattern Detection Techniques for Time-Series Data

Outlier patterns are unusual or surprising patterns that occur rarely, and, thus, have lesser support (frequency of appearance) in the data. Outlier patterns reveal many hidden facts that indicate inconsistency in the data such as fraudulent transactions, network intrusion, change in customer behavior, epidemic and disease severity, intense weather conditions, recession in the economy, etc. Outlier detection has been studied in a variety of data domains including high-dimensional uncertain data, streaming data, network data and time series data. The scope of this survey is limited to time series data. Detecting these outlier patterns rather than other frequent patterns is more important because outlier patterns indicate interesting discrepancies and is crucial for analysis and further decision-making. Outlier values in the data are different from surprising, unusual, or outlier patterns in the data. Outlier Pattern detection in time-ordered sequences discovers in the time series all patterns that exhibit temporal regularities. Considering temporal aspect, interesting outlier patterns can be discovered which otherwise would not have been discovered. The periodicity detection of outlier patterns is to be performed after the detection of these outlier patterns for better analysis of data. Periodic outlier patterns can be found in heart beat pulses, outlier light curves in catalogs of periodic stars, weather data, transactions history, stock price movement, protein and DNA sequences etc. In this paper, the different outlier detection techniques for time series and the existing algorithms in the area are surveyed.