Let’s read in the dataset, updated with the latest information from game week (GW) 8 of season 2019-20.

fpl <- read.csv("https://raw.githubusercontent.com/Jorg3GF/projects/master/gw8.csv")
dim(fpl)
## [1] 551  79

We have 79 variables for 551 players. There are some interesting player attributes here. But I want to make some tweaks. I’d like to have players’ names in a single variable.

Also, I’m used to see their cost in millions, so am going to divide the cost variable by one million.

More importantly, I want to create a new feature. Because we only have 100M pounds of budget, we won’t be able to create a team of “stars” only (all those who score many points each match).

Therefore, an insightful metric to take into consideration is the number of points a given player provides for each penny invested. We call it ROI (Return on Investment) and it goes like this:

\[\begin{equation} ROI = Total Points / Cost \end{equation}\]
fpl$Player <- paste(fpl$FirstName,fpl$Surname)
fpl$Cost <- fpl$Cost/1000000
fpl$ROI <- fpl$TotalPoints/fpl$Cost

Alternatively, we can also see Total Points vs Cost dynamics through the following scatter plot:

We are lucky enough to see that the correlation between cost and total points is not as strong as I thought.

As seen in the correlation matrix above, Total Points and Cost are moderately correlated. In fact, I was also expecting Goals Scored and Assists to record a stronger correlation with Total Points. It seems the most important thing is to play as much as you can - this one I saw it coming! This is what allows us to build up a very strong team with only 100M pounds of budget.

Now, the question is: which sectors should we invest our money in? And what should be the balance between the most expensive players and more affordable ones at each sector of the pitch? Our budget doesn’t allow us to get more than 2 or 3 “stars”, as we still need to get a total of 15 solid players. So, we need to dive deeper in our dataset to uncover them.

I’m curious about the distribution of key metrics across different positions (below I’m considering the top 20 players at each position only).

On average, Goalkeepers (GLK) sum up less points than other positions, as they can only earn them from three main sources, typically: minutes played, clean sheets and saves. In the beginning of this season I picked up a very good one (Ederson) and one of the cheapest (to the bench).

The reasons why I selected Ederson were due to the fact that he plays every single match, his team (Man City) was the team with the highest number of clean sheets and we all know how good he is - so, plenty of beautiful saves ahead.

Turns out that now I am planning to shift my strategy. I want to save money in this sector to allocate it to both the midfield and the attack. Moreover, these first 8 GWs have shown some different dynamics than last season.

My mates claim that this season we won’t see as many clean sheets as in 2018-19. So, how are clean sheets going so far?

Last season, Man City and Liverpool were able to get clean sheets from over 50% of all their 38 matches. So, assuming a linear progression, at GW8 they should be recording 4 clean sheets already. The maximum is 3 though. Still, not a big difference - I do love clean sheets points. But between Man City or Brighton, there’s not a big difference (yet). Probably we can dismiss expensive Man City goalkeepers and defenders. Who knows? Let’s just keep that in mind for now…

So, who have been the top Goalkeepers this season both in terms of Total Points and ROI?

Player Position Team TotalPoints Cost ROI CleanSheets MinutesPlayed
Patricio GLK WOL 38 5.1 7.45 3 720
Mathew Ryan GLK BHA 35 4.6 7.61 3 720
Dean Henderson GLK SHU 33 4.5 7.33 3 720
David de Gea GLK MUN 32 5.6 5.71 2 720
Lukasz Fabianski GLK WHU 32 5.1 6.27 3 573
Nick Pope GLK BUR 32 4.6 6.96 3 720
Vicente Guaita GLK CRY 30 5.0 6.00 3 720
Martin Dubravka GLK NEW 29 5.0 5.80 3 720
Ederson GLK MCI 29 6.1 4.75 3 720
Kasper Schmeichel GLK LEI 28 5.0 5.60 2 720

Rui Patricio is the one scoring more points so far. But Mathew Ryan is cheaper and has performed very well too. So, I am going to go with him! As a substitute, I should go for Henderson or Pope, but I might adopt a different rational. Liverpool’s goalie is as cheap as 4.6M and will play every single match for the top rank team in the Premier League this season. At least until Allison fully recovers.

Before moving onto the next positions, there’s something important we need to take into account. Above, we saw that minutes played emerges as a key element in the game. So, we want to purchase players who play every week. I don’t want to invest my money on a stunning player who will play every two weeks only.

Having said that, we should be aware of those coaches who often change their squad. Therefore, we want to select players from teams where their minutes distribution records a high variance. Who are these teams?

There are two main contenders to win the title this season - Liverpool and Manchester City. So, we tend to think that we should pick 3 players from each one. However, I am getting upset with Man City’s rotation.

Only 20% of Man City players played more than 600 minutes. This number expands to 34% for Liverpool. Therefore, I want to select 3 players from Liverpool for my team, but need to be very conscious when picking up Man City players.

Regarding defenders, what are our choices?

Player Position Team TotalPoints Cost ROI CleanSheets MinutesPlayed
Trent Alexander-Arnold DEF LIV 40 7.2 5.56 2 720
Pereira DEF LEI 40 6.1 6.56 2 720
John Lundstram DEF SHU 39 4.5 8.67 3 688
Patrick van Aanholt DEF CRY 39 5.6 6.96 3 720
Erik Pieters DEF BUR 34 4.8 7.08 2 684
Lewis Dunk DEF BHA 33 4.6 7.17 3 720
Andrew Robertson DEF LIV 33 6.9 4.78 2 720
Enda Stevens DEF SHU 33 5.0 6.60 3 720
Virgil van Dijk DEF LIV 33 6.5 5.08 2 720
Willy Boly DEF WOL 32 5.0 6.40 3 630

In the beginning of the season, my defenders’ selection were Trent-Alexander Arnold, Van Dijk, Laporte, Maguire and Van Aanholt. A very solid defense which provided me loads of points. However, it prevented me from having more expensive players in the attack - the individual minimum cost was 5.5M. As seen above, there’s no need to spend that much. I might keep Trent only, but we’ll see…

Interestingly, defenders are the ones with the highest ROI. I will bear this in mind and keep some money in my pocket for midfielders and forwards - these are the ones who, on average, score the highest number of points.

Concerning midfielders, below there’s a list of good options:

Player Position Team TotalPoints Cost ROI CleanSheets MinutesPlayed
Kevin De Bruyne MID MCI 63 10.0 6.30 3 528
Sadio Mane MID LIV 57 11.7 4.87 3 606
Mohamed Salah MID LIV 55 12.5 4.40 3 708
Raheem Sterling MID MCI 48 12.1 3.97 2 625
Mason Mount MID CHE 46 6.8 6.76 1 709
John McGinn MID AVL 39 5.8 6.72 2 720
James Maddison MID LEI 34 7.1 4.79 1 620
Todd Cantwell MID NOR 33 4.9 6.73 0 685
Richarlison MID EVE 33 7.9 4.18 2 669
Erik Lamela MID TOT 31 6.1 5.08 1 575

I have always played with Salah but I am feeling tempted to let him go and opt for Sadio Mane as he’s cheaper and has summed up a lot of points too. In the long run, Salah might achieve more than Mane, but I need to release some money to get more strong alternatives. I believe will be good to diversify my team and reduce the risk of poor games from top players (yeah, they also experience them - in GW8 Salah only scored 4 points which is very low for the most expensive player of the game).

Finally we have the following options to play in the attack:

Player Position Team TotalPoints Cost ROI CleanSheets MinutesPlayed
Sergio Aguero FWD MCI 61 12.2 5.00 0 560
Pierre-Emerick Aubameyang FWD ARS 55 11.0 5.00 0 720
Tammy Abraham FWD CHE 54 7.7 7.01 0 569
Teemu Pukki FWD NOR 54 7.1 7.61 1 705
Callum Wilson FWD BOU 48 7.9 6.08 1 697
Roberto Firmino FWD LIV 48 9.6 5.00 0 650
Harry Kane FWD TOT 46 10.9 4.22 0 714
Jamie Vardy FWD LEI 46 9.0 5.11 0 720
Wesley FWD AVL 40 6.0 6.67 0 694
Joshua King FWD BOU 38 6.3 6.03 1 708

Since the beginning of the season that I’ve had a cheaper and weaker attack than my mates, though I have ranked at the top position for 5 consecutive game weeks. So, my former strategy of having a solid defense and balanced team was not that bad. Yet I want to try something different now… I want to pick up an astonishing top scorer and see how it goes!

I have developed a Shiny dashboard including not only historical data, but also projections to support all our decisions during the season. Yet, despite all this analysis, one can only win the league if fortune smiles on him/her!