AI Predictions

Poisson match modelling and expected points, computed on the server

ID 1900182 –

Gameweek 20 Predictions

Prev Gameweek 20 Next

Match Outcomes

10 Fixtures

Tuesday 5 January

Arsenal
19:30
Brentford
Brighton
19:30
Bournemouth
Everton
19:30
Aston Villa
Nott'm Forest
19:30
Hull City
Sunderland
20:15
Liverpool

Wednesday 6 January

Crystal Palace
19:30
Chelsea
Fulham
19:30
Spurs
Ipswich Town
19:30
Coventry City
Leeds
20:15
Man City

Thursday 7 January

Man Utd
20:00
Newcastle

Player Projections

GW20 · Top 0
All GKP DEF MID FWD
Any price ≤ £5m ≤ £6.5m ≤ £8m ≤ £10m
Next GW Next 3 GWs Next 5 GWs Most points Best value
No players match these filters.

How did we do in GW10?

Our top 10, as predicted before the deadline, against what they scored

Our top 10 averaged 5.4 points, against 2.9 for the average player who played. The week's top scorer was Barry with 16.

Haaland 6.6 → 2 Gabriel 5.8 → 8 Rice 5.1 → 3 Van Hecke 5.1 → 4 Raya 5.1 → 6 De Cuyper 5.1 → 6 Konsa 5.0 → 8 Evanilson 5.0 → 9 Palmer 5.0 → 2 Verbruggen 4.9 → 6

How these numbers are produced

Match outcomes

Attack and defence ratings are fitted for every club from three seasons of real matches — expected goals, shots on target and the closing betting market — weighted towards recent form and pulled back towards the league average where the evidence is thin. Those set an expected goal rate for both sides, and summing a Poisson grid over every scoreline gives the home / draw / away probabilities, with a correction for how much more often real football ends 0-0 and 1-1 than independence predicts. A club with no top-flight record yet falls back to FPL's own strength ratings.

Player expected points

Per-90 rates for goals, assists, saves, bonus and cards are scored against the real FPL points rules — a clean sheet is the opponent's own goal rate failing to produce a goal, and the conceded deduction is the expected value of the rule rather than half the expected goals. Early in a season those rates are shrunk towards the player's previous season, so one quiet opener does not become his level. That structural estimate is blended with a form prior and with FPL's own next-gameweek estimate, weighted by how much of a record there is to read, and then half-and-half with the player's own points per gameweek so far — the mix that ranked players best when past predictions were scored against what actually happened. Expected minutes lean on the last three gameweeks, so a player who has just been dropped stops looking like a starter.