A Boltzmann theory based dynamic agglomerative hierarchical clustering

In this study, a novel dynamic agglomerative hierarchical clustering algorithm which combines Boltzmann theory of thermodynamics and a graph-theoretic representation of data objects is put forward for data with non-sphere shape clusters. The new algorithm employs neighbors searching operator and vertices spanning operator to construct the linkage paths between vertices. Additionally, in order to obtain the ideal clusters the temperature coefficient is used to completely adjust the linkage paths between vertices. Experimental results on nine benchmark synthetic datasets with different manifold structure demonstrate the effectiveness of the algorithm as a clustering technique. Compared with the K-means algorithm, a genetic algorithm-based clustering algorithm (GAC) and minimum spanning tree clustering algorithm (MST) for clustering task, the presented algorithm has the ability to identify the number and location of the clusters jointly and its clustering performance is clearly better than that of the aforementioned algorithms for complex manifold structures dataset.