PCA used in this way is not particularly helpful without understanding what the principle components actually mean, especially when there's not much structure in the resulting 2D plot.
This highlights a broader point, which is that similar analyses are often performed in scientific research. The real utility of the approach is when it is used to recover latent dimensions that help to explain the differences or clusters in data, and then translate these into meaningful and describable features.
Reading this makes me wonder where game value is these days. I suspect that a curated source of 'decent' Android games would be more valuable for the curation than for the actual games. Of course it would not be easy to extract that value. The two avenues would be charge for use, or charge for entry. And either way the authors are going to wonder if they should get a cut if you're making money off their "free" game.
Saw your spreadsheet. Analysis might have been easier by just ripping open the APKs. For example, City Games LLC "Shoot Bubble Deluxe" is obviously Frozen Bubble code with lingering unused data sprinkled about. (Also the company's domain name was registered once upon a time to an address that was a Kentucky Fried Chicken near Google HQ.)
PCA used in this way is not particularly helpful without understanding what the principle components actually mean, especially when there's not much structure in the resulting 2D plot.
This highlights a broader point, which is that similar analyses are often performed in scientific research. The real utility of the approach is when it is used to recover latent dimensions that help to explain the differences or clusters in data, and then translate these into meaningful and describable features.