Symbolic Aggregate Approximation (SAX)
Kevin Feasel · Apress eBooks · 2022
The prior two chapters introduced a variety of calculations intended to compare two time series and gain an understanding of how far from alignment the two series are. We subsequently adapted this concept to handle multiple series by finding the mean behavior over fairly short segments of approximately seven data points. This is what we implemented as our multi-series analysis technique. In this chapter, we will take a technique that was originally designed to find recurring patterns within a single series and use it to extend our multi-series time series outlier detection engine. Along the way, we will once again create an ensemble of results and use that to inform our scoring of outliers.