An efficient clustering algorithm using graphical techniques

Abstract

This paper aims at an efficient, reliable and scalable density-based clustering algorithm. Using efficient techniques from computational geometry and computer aided geometric design such as the algorithms for range searching, closest pair searching and splines, we are able to derive a clustering algorithm with O(n /spl middot/ logn) in time complexity using O(n /spl middot/ log n/ log log n) in space. In contrast to other density-based algorithms, our algorithm overcomes the use of the parameters for the radius and the minimum number of points in the neighborhood. Comparisons between our algorithm and other known clustering algorithms are provided.