Change point detection

A common problem with time-series is changes in the behavior of the observed system. Generally speaking, a change point signals an abrupt and significant transition between states in the process generating the series. For example, the trend can suddenly change, and a change point can signal where the trend of the series changes. This is well known under the guise of technical chart pattern analysis in trading.

This list captures some applications for Change point detection (CPD):

• Speech recognition: Detection of word and sentence boundaries
• Image analysis: Surveillance on video footage of closed-circuit television
• Fitness: Segmenting human activities over time based on data from motion sensors from smart devices such as watches or phones
• Finance: Identifying changes to trend patterns that could indicate changes from bear to bull markets, or the other way around

As an example for the importance of CPD, consider...