Welcome back to the series! In the previous installment, we dissected the strategies that worked for Week 2 of the 2023 NFL season retroactively using k-means clustering. If you missed it, you can catch up on our initial entry here. Now, armed with insights from the first three weeks, we’re ready to tackle Week 4 and recommend a robust player pool for DFS success.

Methodology (In-Brief)

1. Aggregated projections from six sources for the first four weeks
2. Collected the actual results of the first three weeks.
3. Collected information about the slate including:
   1. Projected Spread: The expected point difference between the two teams set by Vegas.
   2. Game Total: The total expected points to be scored in the game.
   3. Matchup Rating: Based on the average fantasy points allowed to a player’s position by the opposing team.
   4. Salary: We will focus on the DraftKings Main Slate in these.
   5. Value: The standard DFS Metric (projection/salary) * 1000
   6. Risk
   7. Reward
   8. Uncertainty
4. Filtered out only the top 250 players to reduce the player pool.
5. Split out all the players by position.
6. Performed a factor analysis and created a scree plot for each position to determine the appropriate number of clusters.
7. Performed a K-Means Cluster Analysis on each position^[1]If you are unfamiliar with this machine learning technique I would recommend StatQuest on YouTube for an excellent breakdown..
8. Pivoted the Data for visualization.

Clustering Results

I will break down the top two clusters at each position and what the centroid of each looks like so we can identify some patterns for week 4.

QBs

• Chalk is okay at QB.  The top clusters were projected to hit 2.88x and 2.96x respectively going into the week.
• Paying down in the $4,900-$6,000 range has been far more successful so far.
• Analysts have overexaggerated the degree to which bad real-life quarterbacks will be bad for fantasy. The top cluster has the second highest projected risk and the lowest potential reward. We are talking about players like Kenny Pickett and Joshua Dobbs in Week 3.
• Cluster 3 should be your target in Cash.
  • Chalk high projected values
  • low risk
  • high reward
  • cheap

WRs

Cash – You want Cluster 7:

• highest projected multiplier of any group
• lowest risk
• lowest reward (capping your potential for upside)
• higher end salaries ($6,400 to $9,600 in week 4)

GPP – Cluster 1:

• These are your bargain bin players.
• low risk / high reward due to low cost
• some degree of uncertainty in the projection
• This is where you hit on Tank Dell last week.

RBs

• Paying up has been far more successful than looking for value plays like Josh Kelley last week, who only returned a 0.54x multiplier. The lowest cost player in either of these top clusters has been $6,200.
• You want a close matchup or a projected win in the point spread.
• You have two routes to go for risk/reward profiles
  • Cluster 3 should be your cash target with low risk and low reward
  • Cluster 4 should be your target in GPPs with high risk and high reward

TEs

There are two very clear paths at Tight End.

Cluster 5 – Elite tier

• High salary — pay up for players like Travis Kelce or Mark Andrews.
• Low risk.

Cluster 1

• Low cost — $3,200 to $3,800 so far
• Moderate risk
• High reward
• High uncertainty
• Week 3 was a bad week for the position, with just three players hitting at least 3.0x.
• Only Jake Ferguson and Pat Freiermuth cashed from this cluster in Week 3.

DSTs

• Paying up a little bit to at least $3,100 has been successful.
• You want defenses expected to win.
• Low risk
• High reward
• Uncertainty is okay

Week 4 Player Pool

These are the players that fell into the top two or three clusters at each position heading into Week 4. Players noted with a yellow color code in the cluster column should be considered riskier plays for tournament lineups.