Unsupervised Methods for Time-Series
We've discussed forecasting in the previous chapter, and we'll talk about predictions from time-series in the next chapter. The performance of these predictive models is easily undermined by major changes in the data. Recognizing these changes is the domain of unsupervised learning.
In this chapter, we'll describe the specific challenges of unsupervised learning with time-series data. At the core of unsupervised learning is the extraction of structure from time-series, most importantly recognizing similarities between subsequences. This is the essence of anomaly detection (also: outlier detection), where we want to identify sequences that are notably different from the rest of the series.
Time-series data is usually non-stationary, non-linear, and dynamically evolving. An important challenge of working with time-series is recognizing the changes in the underlying processes. This is called change point detection (CPD...
