There Goes the Neighborhood: A Comparison of a "Just-in- Time" SOM with the Traditional Kohonen Map Applied to Segmenting US Medicare Beneficiaries

Variations of the self-organizing map (SOM) paradigm have been introduced that require fewer computational resources such as memory and processing cycles. The familiar SOM invented by T. Kohonen requires iterative selection of a winner node and updating of a neighborhood of nodes in proximity to the winning node. With an application containing thousands of high-dimensional data vectors, developing a SOM can take days running on computer equipment that is typically available to businesses and other organizations. Most organizations dealing with daily decisions need complex information simplified to a few parameters quickly. One of these newer SOM variations can provide quick results and compares favorably with the results produced by a Kohonen SOM. The application example is an analysis of non-institutionalized adults aged 65 and older for the purposes of providing insights on prospective enrollment in a Medicare Part D prescription drug plan and other policy changes for this population.