An improved MST-based clustering for biological data

Graph-based methods are widely used in cluster analysis because of their efficient functionality in a wide variety of problem domains. In this paper, we propose a new clustering algorithm with the help of minimum spanning tree. We remove the inconsistent edges of the MST by defining an error ratio based on the weights of the edges sorted in non-increasing order. Edges are removed until the error ratio is greater than certain value. An important advantage involved in this method is that the same threshold ratio is used for all the data sets. The proposed method is experimented on various two-dimensional synthetic and multi-dimensional real world data sets to prove the efficiency in detecting the complex clusters. Dynamic validity index is used to evaluate the clustering results of multi-dimensional data. The results of the proposed method are also compared with few existing clustering techniques. The results are encouraging.