A Novel Clustering Algorithm Based on Neighborhood Expansion

This paper presents an approach for classification which is based on the neighborhood expansion. The proposed algorithm can (1) find automatically the number of clusters, and (2) classify irregular data set. In the approach, we first defined the distance between a point and a set, then the neighborhood of a data set. The algorithm can begin with any point in the data set and expands the point to a subset of the data set until the subset cannot be expanded again. Next, we can separate the remained subset of the data set in the same way until the correct classification is obtained. The algorithm is easy to control because there are only one parameter i.e. Neighborhood radius need tune. Simulated experiments on data set with different distribution have shown that the algorithm is effective.