PHA: A fast potential-based hierarchical agglomerative clustering method

A novel travel-time based similarity measure for hierarchical clustering

“…The potential field produced by all the data points is computed similarly as in [16]. The total potential at point i is…”

“…It has been shown that each cluster produced this way is a subtree of the edge-weighted tree T [16]. So there is at most one edge connecting any two clusters.…”

“…Because it is very difficult to simulate the movement of mass points using molecular dynamics, many approximations have to be made [10,13,14]. To avoid the complexity in the simulation, potential-based methods have also been proposed [14,15,16]. In the potential-based methods, the clustering results can be derived from the computed gravitational potential field without simulating the data point movements.…”

“…In the potential-based methods, the clustering results can be derived from the computed gravitational potential field without simulating the data point movements. We have proposed a novel potential-based hierarchical clustering method called PHA in one of our previous papers [16]. In the PHA method, the computed potential field and the distance matrix are used to produce an edge-weighted tree of the data points, from which the clustering results are produced efficiently.…”

“…In the PHA method, the computed potential field and the distance matrix are used to produce an edge-weighted tree of the data points, from which the clustering results are produced efficiently. It is shown that the PHA method usually runs much faster and can produce more satisfying results compared to other hierarchical clustering methods [16]. In this work, the travel time instead of the distance between a pair of data points is used to build the edge-weighted tree and to derive the final clustering results.…”