Similar Matches Finder
Enter both sides' points from the table and their recent results, and find the past Premiership matches most like it. What happened in them is the forecast.
Last season's points are over 38 games. Promoted side? Use about 45, what a newly promoted side usually earns.
That makes the gaps for recent form, this season and last season, in points per game.
Each gap is points per game, home side minus away side: over the last five matches, over this season so far, and over last season. Positive means the home side has been better.
k is how many past matches the forecast is built from.
Try an example
How to read it
- Enter what each side has done: points from its last five games, this season so far, and last season. A league table and the last few results have all of it.
- The model only sees the gaps, home side minus away side. It doesn't know who is playing, just how far apart they've been.
- The three chances add up to 100%. A 60% home win still means the home side fails to win four times in ten.
- It knows only results. Injuries, suspensions, transfers and a new manager are invisible to it.
- Change one number at a time to see which ones the model cares about.
- Home win
- Draw
- Away win
The 20 nearest matches
How it works
Each gap is first measured in standard deviations, so the three count the same. Then the distance from your match to every past match is the straight-line gap between their three numbers:
$$d = \sqrt{\Delta_1^2 + \Delta_2^2 + \Delta_3^2}$$In plain football
- \(\Delta_1\) is how far a past match's form gap was from yours, \(\Delta_2\) the same for this season, \(\Delta_3\) for last season. A past match with exactly your gaps has \(d = 0\).
- The forecast is the share of home wins, draws and away wins among the \(k\) nearest past matches, plus one imaginary match at the usual rates so no result is ever impossible.
- The past matches are the 3,900 Premiership matches from 2001/02 to 2020/21. On the five seasons since, \(k = 300\) scored a log loss of 0.949, level with logistic regression.
Drag \(k\) down to 5 and watch the forecast jump around: five matches are mostly luck. That's why the best \(k\) is so large.
Where it goes wrong
- It knows only results. Injuries, suspensions, transfers and new managers are invisible to it.
- It sees gaps, not teams: the nearest matches to a derby can be any two sides that were the same distance apart.
- It treats the three gaps as equally important, although recent form adds almost nothing once the season gaps are known.
- Its past matches come from the Scottish Premiership, where two clubs have usually been far ahead of the rest.
How it was built and tested: Matches like this one. K-nearest neighbours
Results data from football-data.co.uk. These figures come from a statistical model and are for analysis and education. Football remains uncertain and model predictions will frequently be wrong.