Enhanced Data Topology Preservation with Multilevel Interior Growing Self-Organizing Maps

This paper presents a novel architecture of SOM which organizes itself over time. The proposed method called MIGSOM (Multilevel Interior Growing Self-Organizing Maps) which is generated by a growth process. However, the network is a rectangular structure which adds nodes from the boundary as well as the interior of the network. The interior nodes will be added in a superior level of the map. Consequently, MIGSOM can have three-Dimensional structure with multi-levels oriented maps. A performance comparison of three Self -Organizing networks, the Kohonen feature Map (SOM), the Growing Grid (GG) and the proposed MIGSOM is made. For this purpose, the proposed method is tested with synthetic and real datasets. Indeed, we show that our method (MIGSOM) improves better performance for data quantification and topology preservation with similar map size of GG and SOM.