Ubiquitous Resource-Aware Clustering of Data Streams

With the advance of wireless networks and mobile devices, the concept of ubiquitous data mining was proposed. Because mobile devices are resource-constrained, mining data streams with mobile devices poses a great challenge. Therefore, ubiquitous data stream mining has become one of the newest research topics in data mining. Previous research on ubiquitous data stream clustering mainly adopts the AOG approach. Although the AOG approach can continue with mining under a resource-constrained environment, it sacrifices the accuracy of mining results. In this paper, we propose the RA-HCluster algorithm that can be used in mobile devices for clustering stream data. It adapts algorithm settings and compresses stream data based on currently available resources, so that mobile devices can continue with clustering at acceptable accuracy even under low memory resources. Experimental results show that not only is RA-HCluster more accurate than RA-VFKM, it is able to maintain a low and stable memory usage.