This paper analyzes the popular ant-based clustering approach of Lumer/Faieta. Analysis of formulae unveils that ant-based clustering is strongly related to Kohonen's Self-Organizing Batch Map. Known phenomena, e.g. formation of too many and too small clusters, can be explained due to that. Furthermore it is shown how topographic mapping of ant-based methods is substantially improved by means of a modified error function. This is demonstrated on few selected fundamental clustering problems.
展开▼
