In this article we’ll explore the power of k-means clustering analysis to figure out how we could have taken down the slate the most effectively. By isolating the projected Vegas spread, game total, matchup, projected points, salary, risk, and reward, we should be able to uncover some insights to help us in the coming weeks.
Week 2 of the NFL season brought us a plethora of excitement and challenges, especially with Vegas seeing fit to set all the low game totals (I hope you smashed all the overs!) and a lackluster injury ridden Week 1. I delved into this week’s results with a unique approach. I aggregated projections across seven sources, extracted key matchup metrics, and devised a new strategy to dissect the Main Slate retroactively. (Note: while this analysis shows some promise, we are only looking at what worked in Week 2, and therefore we should be cautious when drawing conclusions about future results.)
Understanding the Metrics: Risk and Reward
Before we dive into the analysis, it’s crucial to understand two new metrics we’ve introduced.
Risk: Calculated as (Average Projection – Low Projection)/Player Salary *1000
Reward: Calculated as (High Projection – Average Projection)/Player Salary *1000
Gap: Risk + Reward = The total uncertainty in the player’s projection.
You may be reading these here saying “duh,” but when quantified in this way, they are extremely useful. With them in the standard format for value in DFS you can see rapidly how much you potentially have to lose and how much you stand to gain compared to just an average projected point value.
K-Means Clustering Analysis
To perform our analysis, we employed k-means clustering, a powerful and underutilized machine learning technique. We clustered players based on the following attributes:
- Projected Spread: The expected point difference between the two teams set by Vegas.
- Game Total: The total expected points to be scored in the game.
- Matchup Rating: Based on the average fantasy points allowed to a player’s position by the opposing team.
- Projection: The average points projected by each of the analysts
- Salary: We will focus on the DraftKings Main Slate in these.
- Value: The standard DFS Metric (projection/salary) * 1000
- Risk
- Reward
- Gap
Clustering Insights
In order to make the retrospective analysis more useful I only included players with at least a projected 1.50x multiplier so that only the players in the likely pool would be evaluated.
Running Backs
All you had to do this week to win at the running back position was click sort on the spreadsheet. Sorting by Risk alone:
- The bottom five RBs in the risk metric averaged a 3.28x multiplier with all five cashing. (Cashing here is defined as finishing with at least a 2.0x multiplier.)
- Brian Robinson Jr.
- Christian McCaffrey
- Derrick Henry
- Saquon Barkley
- Joe Mixon (lowest at 2.08x)
- If you expand to the bottom 10, the average multiplier is 2.97x with only A.J. Dillon not cashing.
That second cluster is where your GPP winners are with Kyren Williams and Zack Moss. The issue is that the second cluster is comprised of the extreme risk candidates. The other three players in the cluster averaged 1.21x.
That final cluster is the middle of the pack across all the metrics. This is not where you want to be at the RB position.
