New and Improved Similarity Score
2 min read

New and Improved Similarity Score

Before I had what I called my very simple similarity score, which very simply took the values that were on a radar (just 12 stats) and looked to find the players that had the shortest absolute distance from those spots.

With this new and improved similarity score, I have taken the same basic steps but added a bunch of additional statistics in different categories to get a better overall sense of similarity and where they are most/least similar.

I have broken this into stats categories to try to measure similarity based on certain skills as well as overall. The categories I created were shooting, creating, passing, receiving/carrying, ball-winning (defense), and aerial.

To determine similarity for this I have chosen to use total absolute distance in z-score for each stat. The most similar player is the one with the smallest overall difference in z-scores.

So say we are Liverpool and we need to find a replacement for Mohamed Salah. We can run this and get a shortlist of players who are most similar.

We can also limit things by age, here are the under 26 players that are most similar.

Here is how things look if instead of just looking at similarity score we switch to looking at the crab cake chart.

Salah is a unique player but I think the system did pretty well at identifying players that have a similar statistical profile.

I imagine that this would be fun to help create stats scouting short lists based on a certain profile.