Fraunhofer Institute for Industrial Mathematics ITWMIn the 21st century, huge amounts of data are collected - and many present themselves as so-called integer data. These usually occur when things are counted: How many times is a stock traded per day? How many times is a website accessed per hour? How many different characters do emails contain?
New Insights Through Change Point Detection
By searching for change points, new information about such data can be found, automated and without human bias. In clinical trials, effective drugs should lead to improvement in measured symptoms after the start of intake, but an ineffective drug should not. In case of improvement, the structure of the time series changes after intake and such a point in time is called a change point.
Abrupt and Continuous Changes
An exemplary case for a clinical trial of an integer problem is a drug against seizures in epilepsy patients. In this case, seizures can be expected to decrease from one observation to the next - i.e., abruptly - to a lower level. In other cases, however, a change point may mark a continuous change from one level to the next. When developing case rates of infectious diseases, one can model an outbreak as a change point. However, new techniques are needed to recognize an exponential development as a change point.