The rest of the Premier League season simulated 800 times (27/02).
-
@Arsenal
still favourites for the league
-
@WestHam
looking much better after this weekend
-
@SouthamptonFC
and
@afcbournemouth
relegated 70% of the time
Here's the full table - find the breakdown below.
I have recently taken part in the MCWFC x Hays hackathon – tasked with creating a web-tool for coaches to use, using
@StatsBomb
and
@SecondSpectrum
data.
My team and I created kicksmarter. – a web-tool with two functions:
- Post-match analysis
- In-game observations
The Croatian
#HNL
simulated 1800 times!
-
@NKRijeka
are champions 63% of the time
-
@gnkdinamo
sneak the top spot in 21% of simulations
-
@nkosijek
have a 3% chance of getting European football
-
@NKRudes
are doomed
Here’s the results for my 10,000 EURO 2024 simulations (seen on
@jamesallcott
’s recent video).
- Portugal are favourites, winning almost 15% of the time
- England in 2nd and are champions in 11.5% of simulations
- Scotland with a 50/50 chance of getting out the group
🧵👇
Simulating the rest of the EFL Championship season 2000 times!
-
@IpswichTown
now only have a 13% chance of getting automatic promotion
-
@QPR
with a 66% chance of staying up
-
@HullCity
are in the play-offs in 18% of simulations with a tough run-in
The rest of the Scottish League One season simulated 745 times.
-
@officialdafc
top the table 63.2% of the time
-
@MontroseFC
make the play-offs in 9.1% of simulations
-
@pfcofficial
stay up 25% of the time
Here's the full table - find the breakdown below.
Following on from my results of the Premier League simulation yesterday, here is the same but for the Scottish Premier League.
Simulations are generated using deep neural networks, which use over 10 years of football data to predict results of games.
Back in early March I simulated League Two 2550 times, and Doncaster Rovers did not reach the Play-Offs in a single simulation.
What they have done since then is truly remarkable - full credit to them for absolutely smashing my model's expectations.
@drfc_official
#DRFC
I'm back doing my simulations!
Here's the rest of the Scottish Premier League season simulated 2500 times.
-
@CelticFC
are still favourites for the league
-
@LiviFCOfficial
avoid the automatic drop only 13% of the time
-
@DundeeFC
get European football in 6.5% of simulations
Here's the updated Scottish Premier League simulations!
-
@RangersFC
now (just) favourites for the title
-
@KilmarnockFC
lead the way for the last European spot
-
@StJohnstone
are in the bottom two in 23% of simulations
-
@LiviFCOfficial
giving themselves a chance of survival
The Premier League simulated 4000 times (as seen on
@jamesallcott
's recent video).
-
@LFC
are still favourites
-
@Everton
are relegated only 14% of the time
-
@SpursOfficial
get Top 4 in 59% of simulations
My presentation from the
#StatsBombConference
is now live on YouTube.
Feel free to watch! Any feedback is much appreciated.
Many thanks to
@StatsBomb
for the amazing opportunity - it was a truly surreal experience.
Here's descriptions on how the machine learning simulations take place. I have included a brief shortened version, and a longer in-depth description.
If there are any questions, feel free to ask!
The rest of the Scottish Premier League season simulated 531 times.
-
@Celtic
win the league 96.6% of the time
-
@Aberdeen
reach the Top 6 51.2% of the time
-
@dundeeunitedfc
and
@KilmarnockFC
trading the bottom two spots.
Here's the full table - find the breakdown below.
The grouped breakdown:
I am assuming that 6th place will get the Conference League spot, as Manchester United have won the EFL Cup and will most likely finish in the Top 5.
A rough look at the difficulty of each teams remaining opponents.
Take this with a pinch of salt - this is a just a quick index I created based off of current league positions of opponents (the machine learning will teach itself this).
@Everton
with a tough run-in.
Currently running updated simulations for the Scottish Championship, League One and League Two. Will be ready tonight!
Will be surprised if
@officialdafc
don’t finish top 100% of the time after yesterday…
Was an amazing experience to present kicksmarter. at
@ManCity
today!
It’s even more amazing to win the hackathon - all the presentations were so impressive. The future of football data is in good hands.
Check out kicksmarter. in the tweet below:
#ManCityHaysHack
Inspired by the fantastic
@OptaAnalyst
, here is the result of 103 separate simulations, predicting the final positions of The Premier League.
Simulations are generated using deep neural networks, which use over 10 years of Premier League data to predict results of games.
Here is my model simulating the Premier League season 5300 times!
I'll try and give a quick explanation below on why different models have different probabilities 👇
The rest of the Scottish Championship season simulated 600 times.
-
@queensparkfc
gain automatic promotion 67% of the time
-
@PartickThistle
reach the play-offs in 57% of simulations after yesterdays win
-
@CoveRangersFC
&
@acciesfc
trading 9th place
Find the breakdown below.
Here's descriptions on how the machine learning simulations take place. I have included a brief shortened version, and a longer in-depth description.
If there are any questions, feel free to ask!
I had the idea yesterday to quantify the 'Player Openness' tool that we created last year for the hackathon I took part in.
So here are the results! I'm calling it 'Pitch Tightness' for now - any feedback would be appreciated! I'll describe the process of making this below 👇
Today I'll be simulating a few leagues people have requested:
- Serie B
- Croatian League (HNL)
- Ligue 1
- Undecided (if anyone has suggestions let me know!)
The rest of the SPL season simulated 585 times.
-
@StJohnstone
's top six hopes are slipping
-
@dundeeunitedfc
are automatically relegated 63.8% of the time
-
@AberdeenFC
's chances of Europe have more than doubled since last week
Here's the full table - find the breakdown below.
First I will show-off the post-match analysis. By using player tracking data, we have created a tool which replays a full match from start to finish on 2D tactics board.
This allows coaches to have a simple and digestible overview on player performance and team tactics.
I’m thinking of making a video on the different football predictor models out there and how they compare.
So far I’ve got:
-
@FiveThirtyEight
-
@OptaAnalyst
-
@CannonStats
- My own model
Am I missing any others?
The rest of the
@Bundesliga_EN
season simulated 827 times.
-
@FCBayernEN
are champions in 57.9% of simulations
-
@fcunion_en
make the CL 65.5% of the time
-
@s04_en
with a 75.3% chance of getting relegated
Here's the full table - find the breakdown below.
Just one week left to buy tickets to the 2023
#StatsBombConference
There are a limited number left, going off sale at the end of Friday 6th October
The only conference dedicated to the Pro Analytics industry. Secure your place today:
Secondly, the in-game tool observations gives the coaches an extra pair of eyes, using data live from the game.
A live-feed will make data-led observations every minute, that may be hard to see with the human eye.
The feed posts the most anomalous observation every minute.
The rest of the Scottish Championship, League One and League Two seasons simulated 640 times.
-
@acciesfc
favourite for the drop
-
@officialdafc
promoted in 90% of sims
-
@EastFifeFC
with a 37% chance of reaching the play-offs
Find the grouped breakdown of the positions below.
You can move to any point of the game using a scroll bar, or by clicking on any key event.
Furthermore, there are three extra features:
- A live 'openness' circle showing player space
- The ability to drag-and-drop players
- A simple drawing tool
Here's descriptions on how the machine learning simulations take place. I have included a brief shortened version, and a longer in-depth description.
If there are any questions, feel free to ask!
UPDATE: 19/02
After a weekend of results, along with the
@MotherwellFC
win midweek, here is the updated SPL simulations for the rest of the season, created using Deep Neural Networks (AI).
There are a few significant changes compared to last weeks simulations (see below).
Here's descriptions on how the machine learning simulations take place. I have included a brief shortened version, and a longer in-depth description.
If there are any questions, feel free to ask!
A thread of all my up-to-date simulations. 🧵
Make sure to keep an eye on the date the simulations were made - any matches played since then could alter the results.
If you have any league suggestions, let me know!
A rough look at the difficulty of each teams remaining opponents.
Take this with a pinch of salt - this is a just a quick index I created based off of current league positions of opponents (the machine learning will teach itself this).
The rest of the Premier League season simulated 1000 times - 07/03
-
@Arsenal
are champions 68.1% of the time
-
@OfficialBHAFC
finish in the Top 4 in 19.4% of simulations
-
@LCFC
are relegated in almost 20% of runs.
Here's the full table, find the breakdown below.
The rest of the Ligue 1 season simulated 700 times.
-
@PSG_English
are champions 82.3% of the time
-
@AS_Monaco_EN
and
@RCLens
fighting for that last CL spot
-
@RCSA_English
relegated 31.1% of the time
Here's the full table - find the breakdown below.
Here's descriptions on how the machine learning simulations take place. I have included a brief shortened version, and a longer in-depth description.
If there are any questions, feel free to ask!
@SRFootball_
@CelticFC
@LiviFCOfficial
@DundeeFC
Last year I was in the process of comparing this to the Opta and FiveThiryEight simulations for last season, however got side tracked. I think I remember the models being somewhat similar, however their models were less certain and had a more general spread.
The Polish
#Ekstraklasa
simulated 2535 times!
- 5 teams with a 6%+ chance of winning the title
-
@LegiaWarsawEN
get European football in only 13% of simulations
-
@ZaglebieLubin
's poor form has them slipping into the relegation zone 16% of the time
It's a tight league!
Here's descriptions on how the machine learning simulations take place. I have included a brief shortened version, and a longer in-depth description.
If there are any questions, feel free to ask!
Here's descriptions on how the machine learning simulations take place. I have included a brief shortened version, and a longer in-depth description.
If there are any questions, feel free to ask!
🔵 "The probability is low because the games are running out..."
Michael Beale admits Rangers are running out of time to catch "outstanding" Celtic in title race 🔽
The above results are based on a number of different factors - one being the difficulty of remaining opponents.
I assume that it is this factor that makes the simulations give Bochum and Hertha the edge over Hoffenheim in the relegation battle.
What I'm trying to say, is that we should all models with a pinch of salt - they are just a bit of fun.
They all use different data and methods, and it's interesting to compare them and see why they may differ.
No model can be inherently 'right or wrong'.
@SRFootball_
@CelticFC
@LiviFCOfficial
@DundeeFC
It would be interesting to compare the models! I kept this model simple (no xG, etc) in order to allow me to run it on smaller leagues where I don't have xG data - I would expect the NP xG model to outperform this.
Here's descriptions on how the machine learning simulations take place. I have included a brief shortened version, and a longer in-depth description.
If there are any questions, feel free to ask!
Here's descriptions on how the machine learning simulations take place. I have included a brief shortened version, and a longer in-depth description.
If there are any questions, feel free to ask!
Here's descriptions on how the machine learning simulations take place. I have included a brief shortened version, and a longer in-depth description.
If there are any questions, feel free to ask!
Goal contributions at the World Cup by club (Premier League Top 8):
Manchester United - 10
Tottenham - 7
Chelsea - 4
Manchester City - 4
Arsenal - 2
(Melbourne City - 2)
Newcastle United - 1
Brighton - 1
Liverpool - 0
#WorldCup
Currently running updated La Liga simulations. Spoiler; Barcelona and Madrid are favourites 😮
If anyone has any requests for leagues let me know! The nice thing about this model is that it works on any league out there.
Chucked in this weekends SPL games into my machine learning model, here are its predictions:
Celtic v Aberdeen — 3-0
Hibs v Kilmarnock — 1-1
Dundee United v St Johnstone — 2-2
Livingston v Rangers — 1-1
St Mirren v Ross County — 1-0
Motherwell v Hearts — 0-1
Interesting to see that these results are somewhat similar to mine, even though I use a very alternative method to
@FiveThirtyEight
.
The main difference is their probability for West Ham - this will mostly likely be because they use the SPI index in their simulations.
Who are your tips to beat relegation this season?
Using
@FiveThirtyEight
's model, I looked at each side's form to see who should be worried, and who might have some hope.
For
@TheAthleticFC
:
Commentators again getting it wrong, saying “If Argentina score one more, Poland drop to 3rd”.
This isn’t true, as the goals scored will be equal, as well as the head to head - it then comes down to least number of yellow/red cards, which Poland are winning.
#POLARG
#MEXKSA
Update:
In future models, only one model will be created and used in all simulations. This significantly reduces the time-per-simulation, and gives very similar results.
2. Bournemouth are looking a lot safer - with only a 33% chance of going down compared to 61% last week.
Weirdly, they came 6th in one simulation, by winning 13 of their last 15 games... technically possible, but very very unlikely. Especially with their tough run-in.
Today is the day of
@spfl
. Will be uploading results of simulations for:
- Scottish Premier League
- Scottish Championship
- Scottish League One
- Scottish League Two
As per
@Jon_Mackenzie
's idea a couple days ago, here are two identical simulations of the remaining SPL season with ONE difference:
- The first one uses a model trained solely on SPL data
- The second uses a model trained on EPL and SPL data
'Model' = Deep Neural Network (AI)
A rough look at the difficulty of each teams remaining opponents.
Take this with a pinch of salt - this is a just a quick index I created based off of current league positions of opponents (the machine learning will teach itself this).