Clustering with the Levy Walk: “Hunting” for Clusters

摘要

The Levy Walk (or Levy flight) is a concept fromBiomathematics to describe the hunting–behaviour of manypredatory species. It is a very efficient way to find prey in avery short time frame. We now want to use this concept ina clustering–context to – if you so will – "hunt" for clusters. We describe how we convert this concept into an efficient wayto find cluster centres by linking the data points through thepath the Levy Walk takes. The clusters are then created bystatistical analysis of the links between the data points. The resultis a Clustering algorithm that works on massive datasets withextremely high level of noise. It is not dependent on any form ofcluster shape and is almost free of parameters. The only choiceone has to make is the precision of the Clustering, which in turndetermines the runtime.