On the Window Size for Classification in Changing Environments.

Abstract

Classification in changing environments (commonly known as concept drift) requires adaptation of the classifier to accommodate the changes. One approach is to keep a moving window on the streaming data and constantly update the classifier on it. Here we consider an abrupt change scenario where one set of probability distributions of the classes is instantly replaced with another. For a fixed ‘transition period’ around the change, we derive a generic relationship between the size of the moving window and the classification error rate. We derive expressions for the error in the transition period and for the optimal window size for the case of two Gaussian classes where the concept change is a geometrical displacement of the whole class configuration in the space. A simple window resize strategy based on the derived relationship is proposed and compared with fixed-size windows on a real benchmark data set data set (Electricity Market).