tai rahman

UFC forecast

UFC fight forecast

Choose two UFC fighters and a weight class to see the win probability a forecasting model gives each of them, and what pushed it.

Fighter data
UFC history through October 17, 2026
Model
Model v1 · trained through 2024

The model is frozen; the fighter histories it reads are not. The data is a snapshot, not live, and this isn't betting advice.

The short version

  • Model v1 is a logistic regression built only from what would have been known before each fight, so a forecast for a past fight never uses that fight's result or anything after it.
  • The chart under each forecast ranks the factors by how much they pushed the probability toward one fighter or the other, using the model's own contributions. These are associations the model learned from past fights, not causes.
  • In historical tests, 61.7% of the model's favorites won across 2015 through 2024, and 66.0% won across 2025 and 2026 fights held out from its development. Neither figure is a promise about future fights.
  • It only knows UFC history, so a fighter with a long career outside the UFC can look like a newcomer to it.
1 Choose the fighters

 

 

2 Confirm the fight

Used by the model: rates are smoothed toward this division's averages. Filled in from the fighters' most recent division.

Recorded with the forecast · not used by Model v1

Rounds
Title fight

Choose two fighters.

Methodology

How the forecast works

This forecast comes from Model v1, a logistic regression model trained on UFC fights through 2024. It uses only information that would have been available before the fight: each fighter's UFC history, an Elo rating, striking and takedown statistics, finishing rates, age, reach and time since their last fight. For a new forecast it reads each fighter's UFC history through October 17, 2026.

How the probability works

The model adds up how far each fighter is ahead or behind on each factor, weights each one by what it learned from past fights, and turns the total into a probability. The two probabilities always add to 100%. A 65% forecast means that, across many fights the model rates the same way, the favorite would be expected to lose about a third of them if the model is well calibrated. Scheduled rounds and title status are recorded with a forecast but aren't model inputs, so changing them doesn't change the probability.

How Elo works

Every fighter starts with the same rating. After each UFC fight the winner gains points and the loser loses them, and an upset moves the ratings more than an expected result. Model v1's ratings move in large steps, so the raw numbers look unusual; the page shows only which fighter is rated higher.

Why rates are smoothed

A fighter with two UFC fights might have a 100% knockout rate, which says little. So each rate is blended with the average for the division: the fewer UFC fights a fighter has, the closer their rate stays to that average. Finishing rates, for example, count the division average as if it were ten extra fights.

What the model learned

Using UFC fights from 2005 through 2024, the model learned which pre-fight differences tended to go with winning: a higher Elo rating, more takedowns, being younger, and so on. These are associations. A factor that pushed a forecast toward one fighter is one the model has seen go with winning in the past, not something shown to cause wins.

How the model was evaluated

The 2015 through 2024 figure comes from a walk-forward test: each fight was forecast using only fights before it, as if in real time, and 61.7% of favorites won. The finished model was then scored once on 2025 and 2026 fights it had never seen, where 66.0% won. Both are results on fights that had already happened, so they describe how the model did, not how often it will be right from here. Those 2025 and 2026 fights stay reserved for evaluation, which is why the model is trained only through 2024 even though the fighter histories it reads are newer.

Known limitations
  • The model only knows UFC history. A fighter with a long career elsewhere can look like a UFC debutant to it.
  • UFC 1 is missing from the data, and statistics from the early years are sparse.
  • Height and reach come from fighters' current profiles rather than what was recorded at the time, a small known look-ahead.
  • The fighter data is a fixed snapshot. New fights and new fighters are not added automatically.
  • It is a probability, not a guarantee, and it isn't a betting system.