This is super neat, but the top 5 icons is a very small sample to draw any conclusions from.

If you wrote a script to fetch the icons and do the subsequent analysis, it should be trivial to modify it to fetch a few hundred icons in each category instead, which should yield more accurate results. You may find out, for example, that a few colors are just as equally represented and that there is no real "dominant" color for a category.

But if you fetched the icons by hand, this would be problematic...