Identifying riskier combinations of risky behavior using a self-organizing map

摘要

A variation on the self-organizing map (SOM) introduced by T. Kohonen [1982, 1995] was developed using real-valued, categorical, and binary data in each vector as a tool for multivariate descriptive analysis. Different similarity measures were applied to each type of data and all were combined and normalized to produce a single score as the final similarity measure. The data source is a national telephone survey on health status and behaviors [2001]. One state, New Jersey, was selected for SOM development to both limit the number of vectors and focus on a region of interest. Several nodes and neighbors in the SOM topology revealed combinations of risk that might work synergistically to produce a much higher level of risk. Examples include firearms in the home combined with stress and lack of rest and/or alcohol abuse.