How the prediction models work
Every match analyzed by FUT Análises goes through four independent statistical models. Each one calculates the probability of a home win, a draw and an away win in its own way, and the final prediction is the average of the four. Combining different models reduces the error of any single one.
The four models
Dixon-Coles
Estimates how many goals each team should score from their attacking and defensive strength, with a home-field advantage. It is a Poisson model with a correction for low scores (0-0, 1-0, 0-1 and 1-1), which plain Poisson tends to get wrong. From it come the probabilities for the exact score, goals over or under a line, and both teams to score.
Elo
Gives each team a strength rating that rises when it wins and falls when it loses, in proportion to how surprising the result was. The gap between the two teams' ratings becomes the probability of a win, draw or loss.
Bradley-Terry
Compares teams head to head based on the season's matches and fits a strength value for each one that best explains the observed results. It works as a second opinion to Elo, with a different way of measuring relative strength.
XGBoost
A machine learning model trained on match history. It uses recent form (wins, goals scored and conceded), shots on target, possession, league position, head-to-head history and the outputs of the other models.
What data goes in
Results, match statistics, missing players and standings for the leagues we track, sourced from a professional sports data provider. Every prediction only uses what was known before kickoff: no data from the match itself enters the calculation. Models are recalculated as new matches finish, but the prediction for a finished match is never redone after the result.
How to check
For every finished match we show the result the model pointed to as most likely, and whether it happened, including when it did not. A 60% win prediction means that, across many similar matches, the team would win about 6 out of 10. The other 4 also happen, so a single miss does not say much about the model.
Limits
Football has a lot of variance and few goals per match. The models cannot see last-minute facts, like a lineup change minutes before kickoff, weather or motivation. Cups and the start of a season have less history, so predictions are less precise. This analysis is statistical, informational content for adults 18 and over, not betting advice.