Our methodology
Every prediction starts from a Dixon-Coles double Poisson model fitted on eight seasons of match results across Europe's five major leagues, their second tiers and continental competitions, weighted by exponential time decay (half-life up to two years). A recent-form adjustment and, where coverage exists, per-match expected goals (xG) over each team's last 12 matches refine the goal expectancies for each fixture, a low-score correlation correction (ρ) sharpens the most common scorelines, and the win/draw/loss probabilities are blended with an Elo ratings model for robustness.
We back-test strictly out of sample: the model never sees the matches it is scored on. Headline metrics — top-pick accuracy and Brier score — are printed on every match page.
What we are not
ScoreLens is a data journalism project. We do not accept advertising from gambling operators, we never publish betting tips, and our outputs are statistical estimates for informational and entertainment purposes only. They are not advice of any kind.