← Rankings

Methodology

How the rankings work

How does BOXINGRANK rank boxers?

BOXINGRANK is a statistical model, not a panel of voters. It reads every professional fight in our database, rates each boxer on who they beat, how they beat them and how recently, and publishes pound-for-pound and division rankings that update after every card. The same ratings give a win chance for every booked fight, and every pick is checked in public against the real result.

Where does the data come from?

Results: a historical bout database (originally licensed from boxing-data.com), kept current from the "Professional boxing record" tables on Wikipedia (CC BY-SA 4.0). New results arrive three times a day; a Wikipedia result is marked provisional until it is confirmed or replaced.

Champions come from Wikipedia's lists of current world champions (WBA, WBC, IBF, WBO and The Ring). Punch statistics come from CompuBox and cover televised fights only.

Only professional bouts count: fights scheduled for fewer than four rounds, and decisions before round four, are left out, which keeps amateur and exhibition bouts from leaking in.

How is the rating (0–100) worked out?

Each boxer's rating combines four parts, each scored against every other rated boxer: opposition (48%), form (40%), activity (8%) and stage — title fights and big nights (4%). Ten percent of the final number comes from a Glicko-2 rating, a well-known chess-style rating system that also tracks how sure the model is about each boxer (the ± next to a rating).

Opposition is the average strength of the boxers you beat. Strength is measured within each weight class (how far above or below that division's average an opponent is), so beating a top welterweight counts the same as beating a top heavyweight. That is how a single pound-for-pound list can compare different weight classes.

Form blends a boxer's career win rate (45%) with a recency-weighted win rate (55%, where a fight a year ago counts half as much as one today). Small records are pulled toward the average until a boxer has enough fights to stand on their own, and a boxer whose last fight was a loss takes a 3% deduction.

Underneath, an Elo-style rating moves after every fight. Title fights move it 40% more than other fights, and the size of a win matters: an early knockout counts for more than a late one, and a wide decision more than a split decision.

When two boxers are within 1.5 points and one beat the other in their latest meeting (in the last four years), the winner is placed directly above.

Who gets ranked?

To appear in the rankings a boxer needs at least three rated pro fights and a fight in the last two years. Boxers whose Wikipedia article says they have retired (or who have announced it) drop out unless they have fought in the last 30 days; their profiles stay up.

Men's and women's lists are separate. The pound-for-pound lists (top 50 men, top 30 women) also require a meaningful level of opposition; each division shows its top 15.

How are fight predictions made?

Every booked fight gets a win chance from the two boxers' Glicko-2 ratings and how sure the model is about each. That raw number is then adjusted once, using past fights, so that a boxer the model gives a 70% chance really does win about 70% of the time (this is called calibration). A boxer the model hasn't rated yet gets a neutral starting rating.

The method pick (KO/TKO or decision) comes from the favourite's knockout rate, scaled by how clear a favourite they are. These are our model's estimates, not betting odds, and the site carries no betting content.

How do we know if the model is any good?

The track record page replays every fight in date order and predicts each one using only fights that happened before it — so the model never sees the answer (an “out-of-sample” test) — then checks the pick against the real result. Draws, and fights where either boxer had no earlier rated fight, are left out. It shows how often the favourite won, a Brier score (how close each percentage was to what happened: 0.25 is a coin flip, lower is better) and a table of whether its 60%, 70% and 80% picks win as often as they should. See the current numbers →

What can't the model see?

Boxing records are messy: the same fight can appear twice from different sources (we merge and de-duplicate them), recent results can lag by a few hours, and lightly documented regional cards are thinner than world-title cards.

The model knows results, not context: it can't see injuries, short-notice replacements, controversial scorecards or weight-cut problems. Ratings at the very top are close together, so small rank differences there are within the model's uncertainty.