Resource-Aware Density-and-Grid-Based Clustering in Ubiquitous Data Streams

Because data streams are massive, fast, real-time, and unpredictable, traditional data mining techniques are not suitable for data streams. Hence, data stream mining has attracted much research attention from the data mining community. On the other hand, the increasing processing power of mobile devices has made it possible to run lightweight data mining algorithms on mobile devices. Hence, the concept of ubiquitous data mining has been proposed recently. Because of the constrained resources of mobile devices, however, existing data stream mining algorithms cannot be used on mobile devices and may lead to mining interruption as resources are insufficient. Therefore, in this paper we propose the RA-DCluster al-gorithm that adapts to currently available memory and battery as well as adopts both density-based and grid-based clustering methods so that mobile devices can continue with clustering data streams even under lower memory and battery. Experimental results show that RA-DCluster not only has better stream processing efficiency and mining accuracy than RA-Cluster, but also uses less battery and maintains lower and stable usage of memory.