Nonstationarity describes data or features whose statistical properties change over time or across another dimension. It is important in machine learning because models trained on one distribution may degrade when the real-world distribution shifts.
A feature whose values change across one or more dimensions, usually time. For example, consider the following examples of nonstationarity: - The number of swimsuits sold at a particular store varies with the season. - The quantity of a particular fruit harvested in a particular region is zero for much of the year but large for a brief period. - Due to climate change, annual mean temperatures are shifting. Contrast with stationarity.