Will's Premier League Predictions

[willz121]

Just for a bit of fun, and because I am trying to understand AI / Data Science at work at the moment, I have been looking at using past data to predict the scores for the FMFA Premier League games as we go.

1. All of the results for the Premier League from Season 18 have been collated into the model, providing each team with the following for home and away:

• Average Goals Scored

• Average Goals Conceded

• Attack Power (how likely a team is to score)

• Defence Power (how likely a team is to concede)

2. From this, estimated goals can be calculated for both teams in the fixture:
Home goals = Home Attack Power * Away Defence Power * Home League Average Goals
Away goals = Home Defence Power * Away Attack Power * Away League Average Goals

3. This then gives us a value for both sides, and can be used to predict the likely outcome of the match, and when run through a Poisson Distribution Model (yawn are you bored yet?), we can use it to predict what the actual score will be.

EXAMPLE: Leicester City vs Everton in Round 1
Estimated Goals = 1.10 vs 0.34

When running this through the model, we the most likely result is a 1-0 win, which is expected 26.0% of the time. A 0-0 draw is the next highest predicted result at 23.5%.

The model predicts that the likely outcome of the game is a home win (55.5%), with a draw at 33.2% and an away win at 11.3%.

The likelihood that both teams score is 19.3%, but it is more likely that the game will contain <1.5 goals (57.5% chance).

So if you made it through all of that, good on you because it's not for everyone! I am not expecting this to be that accurate at all due to ESMS being fairly random, T tactics, and other elements outside of our control, but it was a fun way for me to try and develop a new skill!

I'll try to remember to post results here as we go and see how the model performs, and by adding each round of results as it comes, the model could become more accurate... enjoy!

[willz121]

Round 1 - Predictions

Leicester City 1 - 0 Everton (H)
Tottenham Hotspur 1 - 3 Burnley (A)
Aston Villa 1 - 0 Bournemouth (H)
Chelsea 1 - 2 Arsenal (A)
Southampton 1 - 1 Watford (D)
Liverpool 2 - 1 Newcastle (H)

Predictions: 3 Home Wins, 2 Away Wins and 1 Draw

As there is no Premier League result data for the promoted sides, it does skew the predictions slightly, but once more results are added to the model, it will be interesting to see how close the predictions are!

[willz121]

Round 1 - Actuals

I think to record the accuracy of these, we will use David's scoring mechanism for the Predictions League of 40 points for the correct scoreline, and 10 points for the correct outcome (home win / away win / draw)

Leicester_City 2 - 0 Everton (H) = 10 points
Tottenham_Hotspur 1 - 1 Burnley (D) = 0 points
Aston_Villa 0 - 3 Bournemouth (A) = 0 points
Chelsea 0 - 2 Arsenal (A) = 10 points
Southampton 2 - 0 Watford (H) = 0 points
Liverpool 0 - 2 Newcastle_United (A) = 0 points

TOTAL = 20 points

Actuals: 2 Home Wins, 3 Away Wins and 1 Draw

As far as predictions go, they weren't massively close at all with a couple of surprise results in Round 1 of the new season. As predicted, Leicester City did beat Everton, although the model suggested 1-0 would be the more likely result. Spurs stood their ground against a very good Burnley side to start with a point, whilst Bournemouth grabbed a shock result at Villa Park winning 3-0. I predicted a closely fought home win, so to see this swing quite far the other way is quite an upset.

Arsenal managed a clean sheet at Chelsea, and managed to come away with the predicted win. The model had confidence that Chelsea would score, but Arsenal's rearguard stood firm and ensured this wasn't to happen. With the promoted sides, it was difficult to gauge, but Southampton start their new season with a comfortable 2-0 home win over Watford, beating the predicted score draw. Finally, Newcastle United went to Anfield and recorded an excellent 2-0 win to kick of the season, and along with Bournemouth, provide a massive shock at the start of the new season!

Scores

Round 01 = 20 points

[willz121]

Round 2 - Predictions

Tottenham Hotspur 2 - 1 Leicester City (H)
Everton 1 - 1 Aston Villa (D)
Burnley 0 - 0 Chelsea (D)
Bournemouth 1 - 0 Southampton (H)
Arsenal 1 - 2 Liverpool (A)
Watford 1 - 0 Newcastle United (H)

Predictions: 3 Home wins, 2 draws and 1 away win

With limited data for Bournemouth and Southampton going into this game, the model favours a home win, but also the score with the highest percentage was actually a 1-1 draw. As the historical data for both sides builds up, the model should in theory become more accurate. However, as this is football and ESMS, nothing is predictable!

[willz121]

Round 2 - Actuals

Tottenham_Hotspur 0 - 2 Leicester_City (A)
Everton 5 - 2 Aston_Villa (H)
Burnley 1 - 0 Chelsea (H)
Bournemouth 1 - 2 Southampton (A)
Arsenal 0 - 2 Liverpool (A) = 10 points
Watford 0 - 3 Newcastle_United (A)

Total = 10 points

Well this proves why the bookies always win, or that my predictions are actually terrible! Only one predicted outcome was correct this weekend with Liverpool winning at Arsenal. I was close with the score, but Liverpool managed to keep a clean sheet and that meant that whilst the model was close, it wasn't quite perfect to get those 40 points for a correct score.

Running through the results, there was only a 2.6% chance of the outcome being a 2-0 win for Leicester City in this one, but with a different manager at the helm to previous games, it can show how much things can be effected by small movements. Leicester were only 23.6% likely to even win the game, so to win 2-0 was a great result for @bruffio in his first game in charge.

Everton were slight favourites to win this one based on the model, but not enough for it to override the predicted 1-1 scoreline (11.8%). The actual outcome of a 5-2 home win had only a 0.3% chance of occurring based on last seasons results and data so it's very interesting that @Darlington's side managed to kick maths in the balls and got a massive home win.

The game between Burnley and Chelsea was always going to be close, and the model reported a 25% chance of it being a scoreless draw. This was closely followed by a 1-0 win for either side coming in at 17.3%, so it was never going to be a goal fest. Burnley winning was the least likely outcome at 30.6% which shows how close all three outcomes were. A solitary strike won it for the Clarets in the end.

As previously stated, the lack of data for Bournemouth and Southampton going into this fixture made the prediction difficult, and it suggested that Bournemouth win would be most likely (43.0%) but the scoreline would probably be a 1-1 draw (12.4%). With the lack of data, the informed decision was to suggest that Bournemouth would take a 1-0 win (10.6%) and unfortunately that backfired completely as Southampton got a brilliant three points on the road.

Liverpool's 2-0 win at the Emirates was highly predictable based on the model, with the away win coming in at 56.9%. The prediction that the result would be 2-1 was confident, and 3% more likely to happen than Liverpool keeping the clean sheet that they eventually did. Based on previous data, the likelihood that both teams would score was 71.9%, proving just how well Liverpool did to keep a clean sheet against a very good Arsenal side.

Watford were huge favourites here, with a 54.5% chance of winning their home tie against Newcastle United. The visitors themselves had a one in five chance of winning (19.8%) which shows just how big a shock this result was for @harry51's side. In addition to low odds of the away win, a 3-0 result came in at a lowly 0.9% chance, and Watford had a 78.7% chance of scoring at least a goal so a clean sheet is massive for the Magpies!

Scores

Round 01 = 20 points
Round 02 = 10 points (30 points total)

[willz121]

Round 3 - Predictions

I am away from Saturday so will get these posted up now...

Leicester City 1 - 0 Aston Villa (H)
Chelsea 2 - 1 Tottenham Hotspur (H)
Southampton 1 - 0 Everton (H)
Liverpool 1 - 2 Burnley (A)
Newcastle 1 - 1 Bournemouth (D)
Watford 0 - 1 Arsenal (A)

Predictions: 3 Home wins, 2 Away wins and 1 Draw

Leicester City vs Aston Villa was a game I always looked forward to as the Foxes manager as it meant I was facing @Nocky and we've always had a good, friendly rivalry over time. With Tony now in charge, the fixture becomes one of local pride and based on recent results, it looks likely to be a home win. Leicester are 61.4% to win the game, and there is a low propensity for both teams to score (38.4%). In terms of scoring, Leicester have an 80.6% of finding the net compared to Aston Villa's 47.7%, so a 1-0 home win seems the most likely result based on historical data (16.7%).

Chelsea against Tottenham is a very difficult game to predict. Chelsea are more likely to win (44.7%) but both sides are showing a very high propensity to score (85.3% vs 80.6%). Due to this, the most likely result is actually a 1-1 draw at 9.0%, closely followed by a 2-1 home win (8.6%). Based on the percentages being so close for results, I will take the 2-1 home win as a combination of Chelsea being the favourite, BTTS being high and 2-1 not being far behind the 1-1 draw.

Southampton vs Everton is a very weird one to try and predict here as previously stated we have minimal data for Southampton in the Premier League so far. Couple that with Everton's fantastic win over Aston Villa in Round Two, it's strange to predict a 1-0 home win here. However, the statistics all point in favour of Southampton overcoming Everton; Southampton are showing a 57.5% chance of winning the game, and BTTS is only at 29.0%. This is further supported by the likelihood of the teams to score, with Southampton very likely at 73.6% and Everton right at the other end at 39.3%. It all points to a home win, so watch @Darlington prove me completely wrong lol

Burnley's trip to Anfield falls very much in the same category as the Chelsea vs Spurs game. Burnley are the favourite to win (45.4% vs 35.5%) and both teams are likely to score with BTTS coming in at a very high 80.6%. In fact, both teams are around the 90% mark to score at least once, with Liverpool (88.4%) slightly behind Burnley (91.2%) in chance of finding the back of the net. With margin for error, it's likely that Burnley will win this game 2-1 but it's going to be very entertaining between two goalscoring sides!

Newcastle return to St James Park after a fantastic 3-0 win at Watford in the previous round, and it looks as though their form should carry them through to at least another point at home to Bournemouth. Unlike the Watford game though, this one is likely to be very close between the two promoted sides, with the most likely result a 1-1 draw (12.4%), closely followed by a 1-0 home win (10.3%). Whilst the stats point in the direction of a home win, Bournemouth are an excellent side and @Matt will be hoping his side can get something in this away tie.

Finally, Watford host Arsenal and the Hornets will be looking to get back on track after the thrashing they took in Round Two. Arsenal are also looking to bounce back after their defeat at home to Liverpool, and the data is suggesting this will be a closely fought contest. Arsenal are more likely to win (46.1% vs 24.7%) whilst BTTS is low at 40.2% suggesting that the odd goal will win this one. This is supported by the score prediction of a 1-0 away win at 15.6%.

Good luck to all teams when Round Three eventually rolls around!

[willz121]

Round 3 - Actuals

Leicester_City 1 - 0 Aston_Villa (H)
Chelsea 0 - 4 Tottenham_Hotspur (A)
Southampton 0 - 0 Everton (D)
Liverpool 1 - 0 Burnley (A)
Newcastle_United 1 - 3 Bournemouth (D)
Watford 0 - 3 Arsenal (A)

Total = 50 points

My first perfect prediction came for the Leicester City vs Aston Villa fixture, with Leicester securing the predicted 1-0 victory over the Villains. There was such a high chance that Leicester would win and keep a clean sheet based on previous results, and this one actually came off as Villa were unable to improve on their predicted form. I'm still very surprised that I've managed to get one spot on, especially after just three rounds of games!

From perfect to absolutely miles off, with Spurs absolutely battering Chelsea in this one. There isn't even a T sheet to blame for this either, with Spurs stepping up and performing brilliantly against their London rivals. To put it into perspective, there was a 0.9% chance of this result coming off, and even more surprising was that Chelsea didn't score a goal as they had a very high chance of 85% to do so! Spurs will be delighted with the win, and will be looking to build on this form to kick start their season.

I had this down as a narrow home win, and a low chance of Everton scoring. It ended up with a goal less draw, which had a 16% chance of occurring. Once again, this could point to a lack of historical data for Southampton influencing the 1-0 home win prediction. Everton have been lackluster in front of goal away from home historically, but at home they have really turned on the form. Not surprised to get this one wrong, but pleased with the small margin of error.

There was no way I saw either team getting a clean sheet at Anfield in Round 3, but Liverpool managed it. If anything, Burnley were favourites to score, although the percentages all pointed to a very close game. What is interesting is that a 1-0 home win was as low as 2.2% likely to happen, so it shows that Liverpool over performed here based on historical data and secured an excellent win.

Once again, lack of historical data for promoted sides may have skewed this prediction, especially as a 3-1 away win was as low as 3% on the model. Both team scoring was pleasing, but for Bournemouth to score three times in consecutive away games must be very pleasing for the manager. Once this season's data is registered into the model, it should hopefully allow for more accurate predictions for Bournemouth as they've certainly impressed so far this season.

Finally, we are hit here with a T sheet for Watford which gave Arsenal a huge advantage. A 3-0 away win was predicted at 4.1%, in comparison to the 1-0 win for the Gunners at 15.6%. This shows how much of an impact the T sheet can provide over the season, and can skew the model in a way in which is difficult to actually predict. Hopefully with time this should iron out, but makes for an interesting beat for me to consider going forward to see if this can be implemented to the model.

Overall, 50 points this round due to the correct score for the Leicester vs Aston Villa game, and a correct outcome for Arsenal winning away at Watford. This gives me my best performance to date and more than doubles the score from Rounds 1 and 2!

Scores

Round 01 = 20 points
Round 02 = 10 points
Round 03 = 50 points (80 points total)

[willz121]

Round 4 - Predictions

Chelsea 2 - 1 Leicester City (H)
Aston Villa 1 - 0 Southampton (H)
Tottenham Hotspur 0 - 2 Liverpool (A)
Everton 1 - 1 Newcastle (D)
Burnley 0 - 0 Watford (D)
Bournemouth 0 - 1 Arsenal (A)

Predictions: 2 Home wins, 2 Away wins and 2 Draws

Chelsea vs Leicester City

Home win: 	42.8%
Draw:		21.5%
Away win: 	35.7%

Home to score:	85.7%
Away to score:	82.9%
BTTS:		71.0%

This is a really difficult one to call as Chelsea are without a manager at the time of writing, and Leicester have also recently appointed a new manager. With both of these factors in mind, and also the closeness of the prediction model, this game will be a tightly contested affair.

Both sides have above an 80% chance of scoring (Chelsea 85.7% and Leicester 82.9%), whilst the home side have a small advantage when looking at the outcome of the game. There is very little between the 1-1 draw (8.4%) and a 2-1 home win (8.2%), but will stick with the home win due to Chelsea's likelihood to win the tie (42.8%).

Aston Villa vs Southampton

Home win: 	48.9%
Draw:		25.1%
Away win: 	26.0%

Home to score:	79.5%
Away to score:	66.2%
BTTS:		52.6%

Aston Villa take on Southampton with Villa manager Alex Nock increasingly frustrated with his sides recent performances. Once again, there is very little between the two sides, and once again it's a close call between a home win and a score draw. The 1-1 draw is the most likely result at 11.9% but a 1-0 home win is very close at 11.0%. Villa also have a 48.9% chance of winning the game, meaning once again I'll side with the narrow home victory.

Tottenham Hotspur vs Liverpool

Home win: 	10.9%
Draw:		15.4%
Away win: 	73.7%

Home to score:	60.3%
Away to score:	92.8%
BTTS:		56.0%

Well I think that is one of the biggest scores I've seen in terms of predicting the outcome for a match so far, with Liverpool weighing in at 73.7% likely to win the match. Liverpool have a 9.9% chance to win the game 2-0, closely followed at 9.1% for a 2-1 win. The best chance for Spurs is a 2-1 win at 3.2%, although everything points in favour of a straightforward away win for Liverpool... cue ESMS random events to prove me wrong!

Everton vs Newcastle

Home win: 	43.8%
Draw:		23.0%
Away win: 	33.2%

Home to score:	82.6%
Away to score:	77.7%
BTTS:		64.2%

Tricky game to predict at this juncture of the season as Everton's home form massively outweighs their away performances. The most likely scoreline is 1-1 (10.2%) but the next highest two results are a goal either way, with Everton 2-1 at 8.9% or Newcastle 2-1 at 7.6%.

Both teams have a decent propensity to score, although Newcastle have kept two clean sheets in their last two away games. The home win predictor isn't high enough above the other outcomes for me to over rule the 1-1 score prediction here, but I am expecting a close fought game.

Burnley vs Watford

Home win: 	38.4%
Draw:		32.6%
Away win: 	29.0%

Home to score:	62.9%
Away to score:	55.9%
BTTS:		35.2%

Very close affair predicted here, with 0-0 the most likely result at 16.3%, only marginally ahead of a 1-0 home win at 16.2%. There really is very little in it, and with home advantage, I think Burnley could come away with three points. However, the majority of the stats point towards a draw, and I will stick with the predicted stalemate here.

Bournemouth vs Arsenal

Home win: 	17.7%
Draw:		24.7%
Away win: 	57.6%

Home to score:	53.7%
Away to score:	80.1%
BTTS:		43.0%

Based on the data, I would be confident that Arsenal will score here and take three points. The three highest predicted scorelines are 0-1 (14.9%), 0-2 (12.0%) and 1-1 (11.4%) suggesting that if Bournemouth can get a goal, they will have a good chance of getting something from this game.

It's not quite as one sided as some of the other fixtures, but the expectation is certainly leaning towards an Arsenal victory to nil.

[willz121]

Round 4 - Actuals

Chelsea 1 - 2 Leicester_City (A)
Aston_Villa 2 - 2 Southampton (D)
Tottenham_Hotspur 2 - 3 Liverpool (A)
Everton 1 - 3 Newcastle_United (A)
Burnley 1 - 0 Watford (H)
Bournemouth 1 - 0 Arsenal (H)

Total = 10 points

Originally, I predicted that Chelsea would nick this one with their home advantage giving them a slightly higher win ratio than Leicester's away form. Both teams did score as predicted, but the score line was the opposite way round as the Foxes overcame the Blues at the Bridge. The 2-1 away win came in at 7.4% vs the 8.2% home win I predicted, so not massively far off in terms of predictability!

At Villa Park, I again sided with the win percentages when making the prediction, and unfortunately it ended in a score draw. This one goes against the model in a few ways; Firstly, Villa were favourites for the win at 48.9%. Secondly, it was marginal whether both teams would score or not, so for both to score twice is a bit of a curve ball. The 2-2 score line (5.1%) was half as likely as the predicted score (11.0%) or even a 1-1 draw (11.9%), showing that football and maths don't often work together!

As predicted here, Liverpool did indeed get the win that the model was so confident of. At a 73.7% win expectation, it was clear that the away team would come away with the points. However, Spurs didn't roll over for their visitors, and a 3-2 away win had more goals than expected! In fact, the model suggested that a 2-3 result would only happen 3.7% of the time, so a bit of a shock score line, albeit the result was correct.

At Goodison Park, I suggested a 1-1 draw would be the most likely result based on both teams having a high propensity to score goals... so this part was true. The score line however didn't go with the model, as Newcastle turned on the style with a 3-1 away victory. The prediction for this score was 3.8%, which in comparison to the 10.2% prediction for the 1-1 draw was vastly different. Newcastle continue their excellent defensive form away from home, although they couldn't keep a third successive away clean sheet.

There was 0.1% between the 0-0 draw (16.3%) and the actual 1-0 home win (16.2%) and unfortunately I stuck with the higher one and predicted the goal less draw. Burnley got a 1-0 win, and, whilst my prediction was incorrect, it was close!

Arsenal had an 80.1% chance of scoring at least once during this game, and Bournemouth stifled them and got a huge 1-0 home win. In all honesty, I didn't see a home win happening with only a 17.7% chance of the Cherries winning in front of their own fans. A 1-0 home win was less than half the likelihood of the predicted 1-0 away win at 7.1%, but I did say if Bournemouth can get a goal they'd have a great chance of getting something!

A measly 10 points this week for Liverpool winning at Tottenham, but some interesting scores and close calls with the model!

Scores

Round 01 = 20 points
Round 02 = 10 points
Round 03 = 50 points
Round 04 = 10 points (90 points total)

[willz121]

Round 5 - Predictions

Leicester_City 1 - 0 Southampton (H)
Liverpool 1 - 0 Chelsea (H)
Newcastle 1 - 1 Aston_Villa (D)
Watford 1 - 1 Tottenham_Hotspur (D)
Arsenal 2 - 0 Everton (H)
Bournemouth 1 - 2 Burnley (A)

Leicester City 1 - 0 Southampton

Home win: 	53.5%
Draw:		29.0%
Away win: 	17.5%

Home to score:	73.0%
Away to score:	46.0%
BTTS:		33.5%

Leicester City are big favourites to win this match, especially with their visitors looking unlikely to score. However, with Southampton scoring twice in their last two away games, it feels likely that this prediction will be incorrect!

The prediction for a 1-0 home win is 19.1% and is a huge swing in favour of the home side. The best prediction for a Southampton win is 0-1 at 9.0%, 10% lower than that for Leicester to win. A goal less draw is the next highest predicted result at 14.6%, with the model favouring Leicester to win this one!

Liverpool 1 - 0 Chelsea

Home win: 	55.6%
Draw:		24.7%
Away win: 	19.7%

Home to score:	80.3%
Away to score:	57.5%
BTTS:		46.2%

Another 1-0 victory is predicted at Anfield, with a 13.6% prediction for the Reds to win by a solitary goal. Both teams to score is nearer to 50% than the game between Leicester City and Southampton, but Chelsea have a lower propensity to score goals away from home.

Anfield is never an easy place to visit, and I think that their current form will see Liverpool pick up another win here.


Newcastle 1 - 1 Aston Villa

Home win: 	51.8%
Draw:		24.1%
Away win: 	24.1%

Home to score:	81.6%
Away to score:	65.9%
BTTS:		53.8%

Similarly to the last round, the Home win indicator is very high and above 50%, but the most likely scoreline is currently 1-1 (11.4%). This is closely followed by a 1-0 home win at 10.6%.

The odds are fairly stacked in Newcastle's favour based on the model, which is why the 1-1 prediction is odd, but after being proven wrong last time out, I will stick with the highest predicted result and say a 1-1 draw is on the cards at St James Park.

Watford 1 - 1 Tottenham Hotspur

Home win: 	39.4%
Draw:		25.6%
Away win: 	35.0%

Home to score:	75.7%
Away to score:	73.3%
BTTS:		55.5%

This game has draw written all over it based on the outputs of the model. Both sides have a fairly even chance of winning the game, and the 1-1 draw is a 12.1% predicted chance.

With both sides above 70% in their propensity to score, the model seems to predict that there will be over 1.5 goals in the game (75.8%), and an even chance of going above 2.5 goals (51.6%). Once again, I will stick to the model and say another 1-1 draw at Vicarage Road.

Arsenal 2 - 0 Everton

Home win: 	72.4%
Draw:		18.9%
Away win: 	 8.7%

Home to score:	86.5%
Away to score:	41.1%
BTTS:		35.5%

After a shock defeat at Bournemouth in Round 4, Arsenal will be looking to bounce back at the Emirates, and Everton provide an excellent opportunity for three points - according to the model! At 72.4% chance of a home win, Arsenal are massive favourites here, and the prediction for a 1-0 or a 2-0 home win are both even at 16.0%.

I have chosen to go with the 2-0 win purely based on the fact that Over 1.5 goals is 71.9%, and with Everton's low propensity to score away from home (41.1%), it's more likely that Arsenal would grab a couple than both teams scoring.

Bournemouth 1 - 2 Burnley

Home win: 	24.3%
Draw:		20.6%
Away win: 	55.1%

Home to score:	75.0%
Away to score:	88.4%
BTTS:		66.3%

In the last round, I got both Bournemouth and Burnley's scores incorrect, and this game the model is suggesting an away win. The 1-2 score line is predicted at 9.3%, but with recent results, Bournemouth getting a 1-1 draw here is the next highest prediction at 8.7%, suggesting that the model is realising Bournemouth's recent home performances.

There should be goals here, kiss of death lol, with Over 2.5 goals reporting 68.6% suggesting it is more likely than not - enjoy the 0-0 draw chaps!

Good luck to all teams, hope you are enjoying reading this thread?