Improved CURE Clustering Algorithm using Shared Nearest Neighbour Technique

Clustering is the unsupervised learning based grouping of data points based on the similarity between them. Traditional clustering algorithms work well with datasets which have globular/spherical shape. When it comes to non-spherical shaped clusters, they may divide large clusters into small clusters or merge two clusters. CURE(Clustering using REpresentatives) clustering algorithm overcomes the limitations of traditional clustering algorithms for clustering non-globular shaped clusters. CURE algorithm chooses random points as representative points from each cluster and shrinks them towards the centroid of the clusters. The problem arises when data sets do not have a centroid tendency. To resolve this issue, we propose an improved CURE algorithm, where instead of shrinking of representative points, the shared neighbours between the points have been used to form clusters. The representative points which share the same neighbourhood are put together in the same clusters. This allows generating clusters of non-globular shaped clusters. Our neighbourhood based clustering does not get affected by the shape of the clusters. Experimental results of our work demonstrate that CURE clustering using shared nearest neighbours has better performance than CURE