Online Clustering Technique with Adaptable Threshold and Radius for Evolving Data Stream

In today's technological communication age, it has become an emerging challenge to cluster online data streams that are generated continuously and evolve over time. Several methods of data mining have been proposed over the years for managing and clustering these massive amounts of data. Although the variety of fixed radius, buffer based and two-stage-based techniques were developed for the clustering data streams, these are not without their issues. However, this paper presents an online clustering technique with an adaptive threshold and radius for evolving data stream named OCATRED. This technique allows user-defined value both for micro-cluster threshold and radius which outperforms the existing fixed threshold and radius-based algorithms. In this proposed OCATRED, a weight adaptation-based mechanism is developed by analyzing the evolving of data streams which ensures the minimum and maximum threshold value. An additional data counter variable is defined for estimating the size of data that will be inserted in the micro-cluster which mainly secured the purity and accuracy of the cluster. It also eliminates growing macro-cluster which reduces both exceed amount of memory utilization and cluster processing time. Moreover, this adaptive method dynamically defines the threshold and radius value for the volume of the cluster over time which constructs a new arbitrary shape of the micro- cluster. An experiment based on several data sets was performed, and the results were compared to other algorithms for specific purposes. The proposed methods outperformed the current methods, according to the less memory and processing time consumption for the micro-cluster and also more cluster purity and accuracy. As a result, the proposed OCATRED method is more sophisticated and precise.