ClusCTA-MEWMAChart: A new clustering-based technique to detect concept drift in the presence of noise

Many real-world applications generate data streams. Typically, data evolves over time and must be processed on-the-fly without the need for long term storage or reprocessing. In machine learning, the inherent variability or change over time of streams is referred to as concept drift. This phenomenon creates new challenges not present in classical machine learning techniques. In this paper, we present a new clusteringbased technique to detect Concept Drift on data streams, named ClusCTA-MEWMAChart. We compare our algorithm experimentally with 4 different methods to detect concept drift from data streams and determine their robustness in the presence of noisy data. We conducted a set of experiments on synthetic datasets. The results show that the proposed approach has good performance.