How should I compare an FPL forecast with the actual result?

Xiao K · FFTQL developer ·

Save the forecast before its target event, then compare the same object and period with the corresponding actual outcome.

FFTQL data states, timestamps and source policy

Preserve the pre-event version

Record the forecast value, publication time and coverage. After the round, verify the actual result and calculate the relevant error. Note changes in important conditions without rewriting the original prediction. A later recalculated forecast cannot establish what the model said before the outcome.

Choose an evaluation measure that fits the target. Points, rank and probability forecasts represent different outputs and should not all be judged with an unexplained identical formula.

Practical checks

  1. Save the dated pre-deadline forecast.
  2. Match its target to verified actual data and record the error.

Context and sources

All numerical examples are hypothetical. The formulas explain a method, and the linked product pages explain data context. For real analysis, retain the source, units, season, gameweek range and retrieval time so another reader can reproduce it.

Sources and useful pages