Creating health typologies with random forest clustering

In this paper, we describe the creation of a health-specific geodemographic classification system for the whole city of Birmingham UK. Compared to some existing open source and commercial systems, the proposed work has a couple of distinct advantages: (i) It is particularly designed for the public health domain by combining most reliable health data and other sources accounting for the main determinants of health. (ii) A novel random forest clustering algorithm is used for generating clusters and it has several obvious advantages over the commonly used k-means algorithm in practice. These resultant health typologies will help local authorities to understand and design customized health interventions for the population. A Birmingham map illustrating the distribution of all health typologies is produced.