Finding Top-$k$ Anomalous Subsequences in Streaming Time Series Using Adapted HOT SAX

Finding anomalous subsequences in time series is an important problem that exists in many practical applications. If the problem takes place in a streaming background, where a time series is continuously updated over time and the sampling rate of the time series can be very high, it will be a huge challenge for the academic community. The challenge is to solve the problem in the streaming background not only accurately but also efficiently in terms of execution time. To meet this demanding requirement, the paper proposes a method to find out local top-k anomalous subsequences in streaming time series using a well-known algorithm called HOT SAX, which discovers the most anomalous subsequence or discord in time series. Since HOT SAX only works in a static background, the proposed method has adapted HOT SAX to work in the streaming background. In addition, the proposed method supplements data normalization to HOT SAX while finding discords. Experiments on the proposed method demonstrate that the method has fast response and obtains local quality top-k discords in streaming time series.