After reading a bit how categorical data can be considered in clustering, I came to the conclusion that most of the post do not make distinction between nominal scale data e.g. colour: red, green, blue, and ordinal scale data e.g. size: small, medium, large. However, the distances between the items make sense at the ordinal scale even if they are not necessarily the same between all items.
- Can I simply convert ordinal scale data to numeric scale without causing much trouble in clustering? I think yes, for the above reasons, but I would be pleased if you could confirm.
- For nominal scale data, where the distances between the items make no sense, would be harder to capture. The easiest way I have found, if there are not too many items on the scale, was to break down the scale and add a variable for each item. E.g. originally, we have colour: red, green, blue, and we make variable colour_red, clour_green, and colour_blue, where each of them can take a value: 0 or 1. See the post form Jordan A on K-Means clustering for mixed numeric and categorical data. It seems to me a valid numeric scale of type ratio as it has a non-arbitrary zero value expressing the complete absence of something and 1 expressing the presence of something. Do you have experience with this in clustering? Is this a valid approach?