Analytics & odds

BanterBoost’s experimental model uses synced FPL expected goals (xG), expected assists (xA), minutes, starts, player availability and completed match results. The percentages are model estimates, not bookmaker prices.

Goal and assist rates are adjusted for expected playing time and the opponent, with 450 minutes of fixed positional prior data added to reduce the effect of small samples. A Poisson model turns an expected event count into the chance of at least one event: 1 − exp(−rate). Availability adjusts that probability separately. Missing xG or xA stays unknown; zero minutes is not evidence of a 0% chance.

Starts, season minutes and completed team fixtures inform playing-time and 60-minute estimates. Opponent and venue scoring rates use completed results, with five matches of league-average data added for stability. The league starts with priors of 1.5 home goals and 1.2 away goals per match. Team clean sheets use exp(−estimated goals conceded) and require completed scoring evidence for both clubs.

Player clean sheet is an approximation of reaching 60 minutes and the team keeping a full-match clean sheet. It is not the exact probability of earning FPL clean-sheet points. Forwards receive no FPL clean-sheet points, so their player estimate is 0%; their team’s clean-sheet estimate is shown separately.

Choose the next 1, 2, 3 or 5 dated, unstarted matches for each club. A double gameweek counts as two matches and a blank is skipped. Fewer matches may be available near the end of the schedule. Current availability is held in the central forecast; the upper scenario allows recovery in later matches. These are assumptions, not future injury updates. For multiple fixtures, the model estimates the chance of at least one event by combining independent match probabilities. Expected minutes and points are summed. No available fixtures means no estimate.

Expected points combine appearances, goals, assists, clean sheets and the other FPL scoring components shown in each breakdown. Missing required inputs leave the total unknown. The lower and upper scenarios change modelling assumptions; they are not confidence intervals or a promise that the outcome falls inside them.

Lineups, tactical changes and correlated events can change the outcome. Evidence labels describe the available sample; they are not accuracy scores. This model remains experimental. Accuracy reports use saved forecasts made before kickoff and completed results, with the evaluated sample shown alongside the errors.