Champion Ascendant
Research

Predicting the UFC call from the fight record

I wanted to know how much a regional fighter's record actually tells you about whether the UFC calls. So I took 172 headliners from 2024–25, checked every record fight by fight, and trained a model on who got signed.

51%of the fighters the model rated highest got the call
6%of the lowest-rated did (31% overall)
0.70AUC, tested on fighters the model hadn't seen
172fighters, records checked fight by fight

Signed rate by model score

6%
26%
32%
41%
51%
overall 31%
Bottom 20%2nd3rd4thTop 20%
Fighters split into five equal groups by model score. Every fighter was scored by a version of the model trained without them.

How I built it

  1. Fighters

    Headliners from LFA, Cage Warriors, Fury FC, UAE Warriors, Brave CF and OKTAGON, 2024–25.

  2. Records

    Counted from full fight lists on Sherdog, cross-checked on Tapology and promotion sites.

  3. No peeking

    Stats are as of the fight before their listed bout, so nothing from later leaks in.

  4. Model

    Logistic regression (scikit-learn), tested with repeated 5-fold cross-validation.

Checking the records mattered. I found a cancelled fight, a few fighters whose listed bout was actually amateur, and two records missing 4 and 8 wins.

Model vs. the alternatives

Same 172 fighters, three ways of ranking them. AUC = how often a signed fighter ranks above an unsigned one (0.5 is a coin flip).

Coin flip
0.50
Rule of thumb
0.67
range 0.58–0.75
DWCS/UFC ASCENT (hand-built)
0.68
range 0.60–0.76
Machine-learning model
0.70
range 0.62–0.78
Ranges are 95% bootstrap intervals. Rule of thumb = undefeated, 27 or under, finishes 60%+ of wins. The ranges overlap, so the model's edge is small and not proven yet.

What the model weights

These came out of the data, not from me.

Win % and age matter most. Finish rate barely moves it once win % is in. A long win streak actually counts a little against a fighter, probably because long streaks often come against weaker regional opponents.

Win %
+0.70
Wins
+0.22
Has a title win
+0.15
Undefeated
+0.05
Finish rate
+0.05
Total fights
−0.04
Win streak
−0.14
Losses
−0.21
Age
−0.45
Weight per standard deviation. Teal = more likely signed, red = less. With 172 fighters, trust the direction more than the size.

Does the gym matter?

A record doesn't show where someone trains, so I sorted every fighter's gym into elite camp, established team or regional gym (classified with Claude against a fixed set of criteria).

Not really. Elite-camp fighters got signed a bit more, but once the record is in the model, the gym adds nothing.

Fighters with no gym listed went 0 for 11, which looks like a great signal. It isn't. Profiles usually get filled in after someone signs, so a blank profile partly reflects the outcome. I left it out.

Signed rate by gym tier

Elite camp
42%
24 fighters
Established team
25%
24 fighters
Regional gym
34%
113 fighters
No gym listed
0%
11 fighters

Limits

  • 172 is a small sample, so treat every number here as a range.
  • Fighters from late 2025 have had less time to get the call than 2024 fighters, so a few "not signed" will change.
  • Opponent quality, management and what divisions the UFC needs all matter too. None of that is in the model yet.
  • It doesn't predict who wins fights.
Code and data on GitHub How the ASCENT scores work