Optimized Fantasy

Last year (2019–2020) was the first time I tried Fantasy Premier League (top 20%) and Fantasy Bundesliga (top 10k ish). While, not hugely successful, it was fun and I recognize some of the mistakes I made in the first few months. I used different analytic approaches, and some worked for me to identify players on the upswing. Of course, I was never very satisfied with how well they worked.

As I prepare for the coming season, I thought I would try some of the optimization techniques that are available. Many of these rely on Linear Programing to maximize some function, such as Maximize the Number of Points I get. There are plenty of articles on Medium (I used this one: https://medium.com/@joseph.m.oconnor.88/linearly-optimising-fantasy-premier-league-teams-3b76e9694877 ). There is even a website dedicated to the topic: https://fploptimized.com/.

I pretty much just used the basic function supplied on joseph.m.oconnor.88 page. So you can use that if you wish to duplicate my process. The challenge for me is trying to decide on what value to maximize on. I settled on these different choices:

1. Total Points from Previous Season
2. Average Total Points from Previous Seasons
3. Total Points from the first 5 games of the previous season

Total Points from Previous Season

This is probably the most obvious choice, and is what is demonstrated in the linked article above. The problem for me is, you are optimizing a team to receive all the points over an entire season. That really suggests that the team should remain the same over the course of a year. The problem of course with this strategy is that a team never stays that constant: transfers, injuries, loans, highs and lows in performances, etc…

If I use this, the total expected number of points with this squad over a season was 2343.

The squad minus substitutes:

1. Martínez
2. Targett
3. Watkins
4. Dunk
5. Dallas
6. Bamford
7. Harrison
8. Fernandes
9. Kane
10. Son
11. Cresswell

Spot a problem? Maybe a few. For one, I think Kane and Son are excellent players, but they also had a phenomenal year. Can they repeat? The bigger issue here for me is Dallas. I had Dallas the entire season last year (a fluke), but he was listed as a defender who played in the midfield position. He was getting a 6 for a clean sheet and scoring or assisting with goals. This year, he is listed in the midfield and will not get 6, but 1. So the model is really overestimating his potential.

Average Total Points from Previous Seasons

Using a player’s average points over several seasons, you get different results. The model predicts an expected score of 2000.

1. Watkins
2. Azpilicueta
3. Digne
4. Vardy
5. Dallas
6. Meslier
7. Harrison
8. Dias
9. Fernandes
10. Pukki
11. Son

Some improvements here such as including Pukki who did not play in the Premier League last season with Norwich City (relegated and the promoted). But then Dallas and Meslier make an appearance. I had both last year and they turned out to be good picks, but I think there are potentially better options. Neither have selected Salah for the team, opting for Fernandes. I think the model sees the higher points for Fernandes, at .5 million less, and uses the extra money for Vardy and Son.

Total Points from the first 5 games of the previous season

What about basing it off of the performance from the first five games of the last season. The reason for this is because, realistically, I am trying to optimize a team for a few weeks and then adjust as the season progresses. This produces a list of good performances from early in the season and biases the squad. Can we assume they will be back to that form again? I doubt it.

1. Martínez
2. Grealish
3. Mings
4. Zaha
5. James
6. Calvert-Lewin
7. Wilson
8. Kane
9. Son
10. Cresswell
11. Saïss

Improvements

For one, I would like to adjust the points for players that moved from a defensive position to attacking. I also want to remove anyone that is injured from the DataFrame.

When I do this, the expected score based on last season’s total points is 2313.

1. Martínez
2. Targett
3. Bamford
4. Harrison
5. Fernandes
6. Wan-Bissaka
7. Pukki
8. Ward-Prowse
9. Kane
10. Son
11. Cresswell

I would also like to force the use of Salah, so I can reduce the total budget by 12.5 million and 1 midfielder, to see the recommended squad.

The final selection being:

1. Martínez
2. Targett
3. Dunk
4. Bamford
5. Harrison
6. Fernandes
7. Wan-Bissaka
8. Pukki
9. Son
10. Cresswell
11. Soucek
12. Davis
13. Foster
14. Coady

Which, I actually do not think is terrible. Will I go with this team? To be honest, I do not know. There are still a few weeks left before squad selection is due and I would like to see if there are some other ways to maximize the points. For one, it does not take into account the squad difficulty of the first 4 or 5 matches and that may sway me towards on player or another. I think the best approach is going to iteratively pick players I feel confident in and use this to maybe find some cheap alternatives to fill in the positions.