A growing mixed Self-Organizing Map

In data mining field, Self-Organization Map (SOM) has been regarded as an effective data visualization means for presenting the original topological relationship between high-dimensional data on a low-dimensional map. Furthermore, growing SOM (GSOM) was proposed to tackle the fixed map problem via more flexible structures. Whereas, both SOM and GSOM are lack of sufficient ability to appropriately manipulate both numeric and categorical data in the training. Therefore, we propose Growing Mixed SOM (GMixSOM) to integrate distance hierarchy and GSOM, to manipulate mixed-type data and improve the visualized results of GSOM in mixed-type dataset. Experimental results show the proposed method can faithfully present the topological relationship between mixed-type data and provide better spread-out control than GSOM.