CLASSIFICATION OF CONCEPT DRIFT IN EVOLVING DATA STREAM
Mashail Althabiti, Manal Abdullah · 2020
The concept of Data Stream has emerged as a result of the evolution of technologies in different domains such as banking, e-commerce, social media, and many others. It is defined as a sequence of data instances generated at very high speed, which can be hard to store in memory. Thus, it became hard to extract knowledge from the continuous data stream using traditional data mining. Data stream mining algorithms should fulfill some requirements such as limited memory, concept drift detection, and one scan processing. Concept drift must be tracked to avoid poor performance and inaccurate results of predictive models. It refers to changing data stream distributions due to several reasons, including the changes in the environment, individual preferences, or adversary activities. In this chapter, we will present the data stream mining components. The problem of concept drift in classification algorithms and several existing state-of-the-art handling methods are highlighted. Besides, the most used datasets, tools, applications, and evaluation methods will be presented.