Clustering Irregular Data Streams With Fuzzy Induction

The inherent characteristics of data streams necessitate adherence to certain conditions, such as processing data within a single scan due to limitations in space and time, and effectively managing concept evolutions. Clustering algorithms, which have seen various developments over the years for processing fast-paced data streams, have been a popular choice when extracting information from unlabeled data streams. A common assumption among existing data stream clustering algorithms is that such methods require access to uninterrupted data streams from sources (electronic devices), which may not always hold true as devices are subject to malfunctions. As and when these assumptions are violated, existing methods fail to resolve the issues adequately. To address this limitation, a clustering-by-variable method named Clustering Irregular Data Streams (<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">CIDS</i>) has been introduced, focusing on building a robust method capable of addressing irregularities within data streams. These irregularities are handled through fuzzy induction or dissipation in response to changes in the data stream's status quo. Concept evolutions are accommodated within the clustering structure through merge and split operations, guided by statistically significant confirmations. The proposed method's quality and effectiveness have been evaluated against existing approaches using synthetic and real-world datasets. Experimental results demonstrate a high degree of flexibility in handling irregularities and achieving high-quality clustering of data streams.