An online fuzzy model for classification of data streams with drift

Abstract

In this paper, an adaptive fuzzy classifier for online rule learning from real-time data streams is proposed. These kinds of data have some limitations which make them different from batch datasets and therefore the process of learning is confronted with many challenges. Since concept drift is one of the most important challenge among them, different techniques as well as our proposed method focus on solving this issue. Our method sequentially updates the constructed model such that the structure and parameters always remains compatible with any new characteristics of data. For having low computational time of modifying the model, we propose a simple updating formula based on minimizing the classification accuracy in each step through gradient descent. The proposed method achieves results that are better than other fuzzy and non-fuzzy methods.