Adaptive grid-based k-median clustering of streaming data with accuracy guarantee

Data stream clustering has wide applications, such as online financial transactions, telephone records, and network monitoring. Grid-based clustering partitions stream data into cells, derives statistical information of the cells, and then applies clustering on these much smaller statistical information without referring to the input data. Therefore, grid-based clustering is efficient and very suitable for high-throughput data streams, which are continuous, time-varying, and possibly unpredictable. Various grid-based clustering schemes have been proposed. However, to the best of our knowledge, none of them provides an accuracy guarantee for their clustering output. To fill this gap, in this paper we study grid-based k-median clustering. We first develop an accuracy guarantee on the cost difference between grid-based solution and the optimum. Based on the theoretical analysis, we then propose a general and adaptive solution, which partitions stream data into cells of dynamically determined granularity and runs k-median clustering on the statistical information of cells with an accuracy guarantee. An extensive experiment over three real datasets clearly shows that our solution provides high-quality clustering outputs in an efficient way.