Nearest Descent, In-Tree, and Clustering

Clustering aims at discovering the natural groupings in a dataset, prevalent many disciplines that involve multivariate data analysis. In this paper, we propose physically inspired graph-theoretical clustering method, which first makes points organized into an attractive graph, called In-Tree, via rule, Nearest Descent (ND). The rule of ND works to select nearest node descending direction potential as parent each node, is fundamentally different from classical Gradient Descent. constructed In-Tree proves very good candidate for due its particular features and properties. original problem reduced removing inter-cluster edges graph. Pleasingly, those are usually so distinguishable they can be easily determined by automatic edge-cutting methods. We also visualized strategy validate effectiveness experimental results reveal proposed method superior related characteristics cutting methods meaningfulness increasing reliability practice.