Projection accuracy · tested 2026-10-26
Closer to the real points than FPL's own expected points
We tested our projections on 52,571 player-gameweeks from two seasons the model never saw. For the next gameweek they missed the real points by 2.02 on average; FPL's own expected points missed by 2.22. These are the projections the planner uses to pick your transfers, and they go five gameweeks ahead.
Average miss by gameweeks ahead
Root mean square error in FPL points per player per gameweek. Ours = Lineup Optimizer. FPL xP = FPL's own expected points for that same gameweek. Lower is better; the best in each row is green.
| GWs ahead | Ours | FPL xP | Form | Pts/game | FPL xP, week before |
|---|---|---|---|---|---|
| 1 | 2.02 | 2.22 | 2.17 | 2.27 | 2.48 |
| 2 | 2.07 | 2.23 | 2.23 | 2.28 | 2.56 |
| 3 | 2.11 | 2.25 | 2.28 | 2.29 | 2.58 |
| 4 | 2.14 | 2.26 | 2.30 | 2.30 | 2.58 |
| 5 | 2.16 | 2.26 | 2.32 | 2.31 | 2.59 |
How we tested it
For every gameweek from GW4 onwards in 2024-25 and 2025-26, we rebuilt the model using only data from before that deadline and predicted the next five gameweeks for every player. Then we compared each prediction with the points the player actually scored. The model's settings were chosen on 2023-24 only, so the test seasons are a fair check. We rerun the test when the model changes; this run is from 2026-10-26.
The model predicts minutes, goals and assists from expected goals and assists (xG and xA), clean sheets from team strength and the opponent, saves, bonus and defensive contribution points. Injury and suspension flags from FPL are applied every three hours.
Questions
How accurate are Lineup Optimizer projections?
Tested on 52,571 player-gameweeks from the 2024-25 and 2025-26 seasons, our projections for the next gameweek missed the real points by 2.02 on average (root mean square error), against 2.22 for FPL's own expected points.
Was the model tuned on the seasons it was tested on?
No. The settings were tuned on 2023-24 only. Every projection in the test used only data from before that gameweek's deadline.
Is FPL's expected points a fair comparison?
It is the number most managers look at. FPL publishes it for the next gameweek only, and the historical values we compare with were recorded around the match, so if anything they had more information than our forecast did.
Does this mean you beat every FPL tool?
No, and we do not claim that. Other tools do not publish their history, so they cannot be compared like this. We publish ours so you can judge it.
Where does the data come from?
Historical points and FPL expected points from the vaastav/Fantasy-Premier-League dataset (MIT licence), and live data from FPL's public API.