Geodemographic model variable selection spacial data mining of the 2011 Irish census

Building geodemographic models using census data is not new. However, many of the models are built commercially and often little is published about how the models are designed and created. One of the important steps to building any geodemographic system is identifying the variables that will produce meaningful and application-relevant clusters. Unfortunately many of the commonly used methodologies for feature selection cannot be used on the census data and so were not available for clustering tasks. This resulted in a more objective-focused approach being taken when choosing variables. This paper outlines the reasons more commonly-used feature selection techniques could not be used, and then explains the alternative approaches and methodologies that were used to select variables to help build an Irish geodemographic model.