Review of Adaptive Data Stream Classification

Data stream classification is a method of mining knowledge from continuous data points. It is classification and prediction task for evolving data streams. For a non-stationary dataset, the data stream classification is posed with the number of challenges like concept drift, infinite length, concept evolution and feature evolution. Data stream is an unending flow of data, which is generated continuously at a rapid rate. As data streams are of infinite length, traditional multi-pass learning algorithms are not applicable as they may require large amount of storage space and training time. Concept drift arrives when the class definition of some instances changes with time. Concept evolution is emergence of new class as stream progresses. However, it is possible that both concept drift and concept evolution may arrive at the same time. By considering these problems, it is challenging to learn a classification model that is consistent with the current concept. Feature evolution occurs when feature space changes with new stream instances, then the feature space of classification model and new unlabelled data would be different, which affects classification accuracy. This paper discusses the different approaches to solve the issues in data stream classification.