StreamSVC: A New Approach To Cluster Large And High-Dimensional Data Streams

The data stream mining has been studied extensively in recent years. This paper is introducing a novel method to cluster high-dimensional data streams, based on famous SVC method, named StreamSVC. SVC projects the images of the data points in a high dimensional feature space, to search for the minimal enclosing sphere, then classifies the points with respect to the distance between each point's image and the central of feature sphere. In StreamSVC, for a single change in the data stream environment, the algorithm redoes the classification part. The algorithm involves only the parts of the data set which are affected during the change of stream and updates the classes in an appropriate time complexity order. Also, in order to update the clusters, in the stream process, we used some new improvements in the labeling piece of original SVC. These improvements are applied to reduce the computational costs for classification part and the cluster's labeling piece. The experimental results show both time efficiency and high accuracy for large data streams.