Cluster analysis using the GSOM: Patterns in epidemiology

The growing self organizing map (GSOM) has proven Itself as an adaptive and robust knowledge discovery algorithm with applications in diverse disciplines. The unrestricted network topology of the learning phase of GSOM enables clustering of data vectors based on the notion of similarity. The hierarchical clustering capability of the GSOM allows drill down examination of clusters at different levels of correlation. In addition to static patterns in the form of cluster groups, the feature map can also contain dynamic information of the dataset. There is continuous experimenting on enhancing the functionality of GSOM as a comprehensive data analysis and knowledge discovery application. This paper documents such extensions along with their application in knowledge discovery. The extensions were applied to a population based sample of children born to families in which either parents or/and siblings have had asthma, eczema or hay fever. The dataset contained time variant information on rash development and allergy diagnosis by skin prick testing to different allergens.