How it works

Predicta is not a crystal ball. It is a classic statistical model adapted to football, and every step can be explained simply. Here is what happens when you ask for an estimate.

  1. 1. Fetch the matches

    The app asks the Predicta server for both teams’ latest results and the league’s average goals. The data comes from a football data API; the access key stays on the server and is never in the app.

  2. 2. Measure attack and defense

    For each team, goals scored and conceded are compared with the league average. A team that scores more than average has a strong attack; a team that concedes less has a solid defense.

  3. 3. Give more weight to recent games

    A match played last week says more about current form than one from three months ago. Each older match therefore counts slightly less than the next one.

  4. 4. Avoid extreme values

    With few matches, an exceptional run can be misleading. The model therefore pulls values toward the average, more strongly the smaller the sample.

  5. 5. Account for the venue

    Home advantage is not a fixed bonus: it is derived from the average goals actually observed at home and away in the league.

  6. 6. Calculate the probabilities

    Each team’s expected goals feed a Poisson distribution. From it we derive the probability of each scoreline, then of a win, a draw and a defeat. A Dixon-Coles correction adjusts low scores (0-0, 1-0, 0-1, 1-1), which a simple model describes poorly.

  7. 7. Head-to-head meetings

    If you choose this basis, matches between the two teams are blended with their current form. They are never used on their own: the sample is too small and squads change.

  8. 8. Check afterwards

    Once the match has been played, the app compares the estimate with the final score. The main measure is the correct outcome (win, draw or defeat); the exact score is a secondary measure, and a much harder one.

What the model doesn’t know

  • Injuries, suspensions and team line-ups are not taken into account.
  • Motivation, weather, managerial changes and what is at stake in the match are not modeled.
  • Few available matches (early season, newly promoted team) make the estimate less reliable.
  • A 60% probability means the event fails to happen four times out of ten: it does happen regularly.

To go further, read our blog articles.