Clustering of data streams based on the history of clusters evolution
Clustering of data streams based on the history of clusters evolution
Jamileh Hosseini1 Adel Ghazi Khani2
1) Iran Imam Reza International University,Computer Engineering-Software,
2) Iran Imam Reza International University,Computer Engineering
IOI Code
:
XBKF-EBBZB
Publication :
Third National Conference on Electrical and Computer Engineering(iscelec.ir)
Abstract :
A data stream is an infinite sequence of data that is continuously being generated. They are collected and analyzed from real-time systems. Clustering is one of the most important unsupervised learning problems. It is one of the most common data mining techniques that aims to find similar clusters among data entry samples. The k_means algorithm is also one of the most popular clustering algorithms which, despite its many advantages, has a significant limitation, which is to specify the number of clusters(k). In k_means algorithms, most of the time, the number of clusters (k) is already known, which is one of the significant weaknesses of this algorithm. Evolutionary algorithms are random search procedures using genetic mechanisms and natural selection. FEAC_Stream is a fast evolutionary algorithm for clustering data streams that automatically allows K estimation of data online. In this study, we improved a fast evolutionary algorithm by using cluster evolution history to cluster the data streams. In this way, we used a buffer to improve accuracy. We used three real datasets to evaluate the proposed algorithm, and we measured the level of improved accuracy that is the same as ss criterion.Consequently, this analysis can be useful for understanding the behavior of each cluster from the beginning of the data stream.
Keywords : data stream clustering evolutionary algorithms concept drift data stream clustering k means