How it works
Several models each estimate how many goals both teams will score. Their blend, adjusted for injuries, rest and travel, gives the chance of every scoreline. The win, draw and loss probabilities and every other number on the site come from those scoreline chances.
Data
| Ligat HaAl | Sport5 | results since 2016/17, fixtures, Hebrew names |
| Premier League | football-data.co.uk | results since 2005/06, shots, xG, bookmaker odds |
| Premier League team news | Fantasy Premier League | official injury status and chance of playing |
| Team news, all other matches | 365Scores | missing and doubtful players, with the reason |
| Champions, Europa & Conference League | UEFA | results since 2015/16, including qualifiers |
| National teams | UEFA and an open archive of internationals | every men's international since 1990 |
| Weather | Open-Meteo | forecast at kickoff |
| News | Israeli sports sites and Google News | Hebrew and English headlines |
Team ratings
An Elo rating for every club, updated after each match. All clubs are rated on one scale, so European results show how strong Israeli teams are compared with English ones. National teams have their own ratings.
Attack and defence
A goals model gives each team an attack and a defence strength, with recent matches counting most. Teams with few games lean on their league's average until the evidence builds up. A second version also uses expected goals (xG), which is steadier than goals alone.
Betting market
For Premier League matches, bookmaker odds are turned into expected goals and weigh heavily, because they are hard to beat. The site also keeps an odds-free prediction and flags matches where the two disagree.
Tested before it is trusted
The blend was tuned on past seasons as if predicting them live, one week at a time, never using later results. Its accuracy is then measured on a separate, later period. The Track record page shows those results and every prediction made before kickoff since.
What adjusts a prediction
- Rest and travel: short turnarounds and long away trips, where past results show a real effect.
- Two-legged ties: the first-leg score changes how both teams approach the second leg.
- Missing players: each absence counts in proportion to how often the player starts (or, in the Premier League, his share of the team's chances). Absences come from the official injury list for the Premier League and from 365Scores' team news elsewhere. Until a match's list is published, a language model running on our own machine reads the Hebrew and English news for injuries, suspensions and returns. The size of each effect was checked against two seasons of lineups (see Track record).
- Weather: heavy rain or strong wind lowers the expected number of goals a little.
Season and tie simulations
The rest of each league season is played out 20,000 times to estimate title, Europe and relegation chances, including Ligat HaAl's split into top and bottom groups. Two-legged European ties are simulated through extra time and penalties.
Limitations
- Team news lists are published in the days before a match; earlier than that, absences come from the press.
- Domestic cup matches are not included, so rest after a cup tie can be overstated.
- Newly promoted clubs start from an estimate until they have played a few games.
- These are probabilities, not certainties: a 60% favourite still fails to win 4 times in 10.