Abstract: The topic of Change Point Detection (CPD) has become more and more relevant as time series datasets increase in size and often contain repeated patterns. By detecting change points in data, segmentation can be performed to group different phases of the time series together. This is of importance for datasets where multiple phases are present and need to be separated in order to be compared. To detect the points, a number of algorithms have been developed and are based on different principles. One approach is to define an optimisation problem and minimise a cost function along with a penalty function. The other approach uses Bayesian statistics to predict the probability of a change point at a specific time. The performance of algorithm and approach can vary depending on the data at hand. This thesis explores how the mentioned approaches are affected by traits in the data. For a firm application link, real world datasets, provided by ABB, are explored in the study.
Handledare: Larisa Beilina