Identifying NBA Player Archetypes Using KMeans Clustering - Part Two
#DFS #DailyFantasySports
Check me out on Patreon!
/ nicksniche
Today we are going to be applying the K-Means Clustering Algorithm to identify player archetypes to better identify performance trends in draftkings or fanduel daily fantasy sports. Seeing the points a team gives up to a position on a nightly basis helps, but there is a lot of variance of playstyle from position to position. We can use machine learning to better drill down and see how certain playstyles perform against certain teams for a more accurate prediction.
Now that we have determined the 'ideal' number of components, we need to determine the ideal number of clusters. This is something we recently discussed briefly in our last intro to K-means video.
We are going to be using an alternative method of determining the 'optimal' number of clusters than we used previously. Today we are going to be using the silhouette method.
The silhouette method is basically a calculation that demonstrates both how similar a point is to other points in its own cluster, and how dissimilar a point is to points in other clusters.
A silhouette score will be between -1 and 1, with higher the score, the better.
If you are mathematically inclined, or just like to know how things work, check out the Wikipedia page for silhouette score to see the actual math behind it, I just don't think it's necessary to get into that level to understand what it does and why it's valuable.
One more important to note here is that more is not always better. And even if it is better, the diminishing returns may not be worth the granularity.
In order to better see the advantages/disadvantages we will be viewing this data relative to the cluster number before it. So instead of looking at usefulness, we want to know where the biggest jumps in value are.
As we can see, the largest percentage increase is going to be from cluster 12 to cluster 13, so for the purposes of this tutorial we are going to use 13 clusters.
Now we simply need to run the KMeans clustering algorithm with 13 clusters and review the results!
Here's my first attempt at classifying these clusters, let me know down in comments what you think each cluster should be classified as!
0. Defensive Ball Handlers / Scoring in paint ball handlers
1. assortment of low minute bench wings
2. defensive / rebounding wing/bigs
3. Offensive Big Men / Offensive Rebounding and Paint Scoring
4. Ball Dominant Stars
5. Rim Protectors / Rim Runners
6. 3 and D Wings
7. Ball Dominant Shooters/Scorers
8. Paul Watson(?)
9. Scoring Wings / Not All-Stars
10. No Clue, Not Importants
11. Defensive Anchor big men?? Not really sure here
12.Bench Scorers
Help me help you! -- Affiliate Links
Patreon -- Help support the channel and get priority access to me for any questions, as well as access to the private discord with other members to discuss dfs in general!
Patreon - www.patreon.com/nicksniche
Solace Bands -- Need an Apple Watch band? Check out Solace bands. I have used them for a while now and they've recently welcomed me into their creator affiliate club so I can get y'all 10% off your order!
SolaceBands affiliate link - www.solacebands.com/nicksniche
SolaceBands Discount Code - NICKSNICHE - Use at checkout
Club EarlyBird -- I don't know about you, but I often times have trouble getting out of bed in the morning, and also staying asleep at night. Talk about a Lose-Lose. Until I found and tried this Earlybird concoction. In the morning it's some of the cleanest caffeine/energy I've ever experienced, and the nightcap keeps me asleep all night and waking up feeling more refreshed than ever. Highly recommend giving it a try with or without my affiliate link!
Club EarlyBird affiliate Link -- Use for 10% off -- https://club-earlybird.myshopify.com/...