Growing Self-Organizing Map with cross insert for mixed-type data clustering

Self-Organizing Map (SOM) possesses effective capability for visualizing high-dimensional data. Therefore, SOM has numerous applications in visualized clustering. Many growing SOMs have been proposed to overcome the constraint of having a fixed map size in conventional SOMs. However, most growing SOMs lack a robust solution to process mixed-type data which may include numeric, ordinal and categorical values in a dataset. Moreover, the growing scheme has an impact on the quality of resultant maps. In this paper, we propose a Growing Mixed-type SOM (GMixSOM), combining a value representation mechanism distance hierarchy with a novel growing scheme to tackle the problem of analyzing mixed-type data and to improve the quality of the projection map. Experimental results on synthetic and real-world datasets demonstrate that the proposed mechanism is feasible and the growing scheme yields better projection maps than the existing method.

Figure optionsDownload as PowerPoint slideHighlights
► Semantic similarity inherent in categorical values can be considered during training.
► The scheme also unifies the representation of numeric, ordinal and nominal values.
► A new neuron-insertion method is devised which prevents generating redundant neurons during growing.
► When applied to data cluster analysis, GMixSOM facilitates obtaining better clustering result.
► Similarity inherent in categorical values is reflected in the projection maps.