Statistical σ-partition clustering over data streams

Abstract

This paper proposes a grid-based clustering method that dynamically partitions the range of a grid-cell based on its distribution statistics of data elements in a data stream. Initially the multi-dimensional space of a data domain is partitioned into a set of mutually exclusive equal-size initial cells. As a new data element is generated continuously, each cell monitors the distribution statistics of data elements within its range. When the support of data elements in a cell becomes high enough, the cell is dynamically divided into two mutually exclusive smaller cells called intermediate cells by assuming the distribution of data elements is a normal distribution. Eventually, the dense sub-range of an initial cell is recursively partitioned until it becomes the smallest cell called a unit cell. In order to minimize the number of cells, a sparse intermediate or unit cell can be pruned if its support becomes much less than a minimum support. The performance of the proposed method is comparatively analyzed through a series of experiments.