Combining SOM and local minimum enclosing spheres for novelty detection

In this paper, a novelty detection method based on self-organizing map (SOM) and local minimum enclosing spheres is proposed. There are two phases in the proposed approach. In the first phase, the whole training set are split into disjointed Voronoi regions by SOM. In the second phase, several local minimum enclosing spheres are constructed upon these Voronoi regions. Compared with its related works, the proposed method demonstrates better performances on one synthetic data set and two benchmark data sets.