Impact of data encoding and thyroids dysfunctions.

In medical environments it is usual to previously encode some variables following medical criteria. In the context of thyroids dysfunctions, the levels of the hormones T3, T4 or TSH are usually treated as low, normal or high instead of using their numerical form, and this is a very common practice in other items. In medical environments it is also frequent to need a clustering process to analyze data. Clus-tering algorithms use always some distance or similarity coefficient to decide which elements have to be grouped in a cluster. The nature of the data determines which distance/similarity can be used, and can change results. In this paper, the impact of preprocessing numerical levels of hormones into categorical labels for clustering is studied, by means of a real sample of patients from a Hospital in Zagreb (Croatia).