Unsupervised Machine Learning to Cluster League of Legends Champions

Hello everyone.

I made the following study, "Unsupervised Machine Learning to Cluster League of Legends Champions", as an exercise to apply machine learning to a hobby. https://i.redd.it/g893bm1pgop11.jpg

The idea was to confirm the existence of the Champion Classes as proposed by riot in this two blogs blog 1, blog 2, by clustering the Champions and seeing if the resulting clusters resemble the classes.

Although it was not an extensive study, I found that almost all classes have a similarity with the resulting clusters, except for the Fighters and Slayers.

Instead of the Slayers, I found a cluster with the following characteristics: https://i.redd.it/r4jo11yjhop11.jpg

The champions in this cluster are like those that we can find in the Slayer class plus the Divers subclass. On average, these champions have high damage and mobility, a medium control and toughness, and they weakness are only a low attack range and low utility, which are easily compensated with their strengths. So, from my point of view, the champions in this cluster are unbalanced, having high kill potential with not so clear weaknesses.

Instead of Fighters, I found a cluster with the following characteristics: https://i.redd.it/z7c9sw75iop11.jpg

The champions in this cluster are like those that we can find in the Juggernaut subclass. On average, these champions have a high base attack damage and toughness, but also they have high control. I also consider that the champions on this cluster are unbalanced since they basically are tanks without the disadvantage of a low damage.

I really enjoyed doing this study, but what I like the most is knowing that studies like this could help to identify balance problems in the game and therefore help to improve it.

P.S. I tried my best to translate the study, but since English is not my native language, I apologize in advance for any errors it could have.