How It Works
The full methodology behind Fight Detective's AI prediction engine
The Prediction Pipeline
Data Collection
Pull fighter stats, fight results, and odds from multiple sources. Every data point is time-stamped to prevent future information leaking into past predictions.
Feature Engineering
Transform raw data into 50+ predictive features. Style matchup scores, rolling form averages, Elo trajectories, and physical advantage composites — all computed point-in-time.
Model Training
Gradient boosting ensemble (XGBoost + HistGradientBoosting) trained with recency weighting — recent fights count more. Walk-forward validation ensures honest accuracy reporting.
Bet Signal Classification
Each prediction is classified into Strong Bet, Bet, Lean, or Skip based on model confidence, edge size vs Vegas, and whether the model disagrees with the favorite. Rules derived from 940-fight backtest.
Prediction & Logging
Predictions are generated and logged before each fight night. Every call is public — wins and losses. No edits, no hindsight, no exceptions.
What Goes In
6,000+ Historical Fights
Every UFC fight in the modern era. Outcomes, methods, rounds, referee stoppages — all structured and time-stamped.
Fighter Statistics
Significant strikes landed/absorbed per minute, takedown accuracy, submission attempts, striking defense, control time, and more.
Elo Ratings & Momentum
Dynamic Elo ratings that update after every fight. Momentum tracking across win/loss streaks and finish sequences.
Style Matchup Analysis
Wrestler vs striker scores, grappling exploit metrics, takedown offense vs defense matchups, striking accuracy differentials, and head-hunting tendencies.
50+ Engineered Features
Raw stats alone don't predict fights. The model uses engineered features that capture how fighters match up against each other — not just how good they are in isolation.
Core Ratings
- ▸Elo differential
- ▸Layoff differential
- ▸Form differential
- ▸Height/reach/age differentials
- ▸Win/loss streaks
Fight Stats
- ▸Striking accuracy
- ▸Takedown accuracy
- ▸Submission attempt rate
- ▸Control time average
- ▸Head strike percentage
- ▸Takedown defense
Style Matchups
- ▸Wrestler vs striker score
- ▸Grapple exploit score
- ▸Strike exploit score
- ▸Ground threat differential
- ▸Pressure score
- ▸Stance mismatch
Recent Form
- ▸Last-5 win rate
- ▸Recent finish rate
- ▸Recent striking volume
- ▸Recent takedown rate
- ▸Recent control time
Career Context
- ▸Strength of schedule
- ▸Quality win rate
- ▸Title bout experience
- ▸Experience mismatch
- ▸Ring rust indicators
Trajectory
- ▸Elo momentum
- ▸Consecutive finishes
- ▸Days since last loss
- ▸Career decline signals
No Hindsight. No Rewrites.
Every prediction is logged to our database before the fight happens. We publish the misses alongside the wins. The accuracy numbers you see are from walk-forward testing — the model was tested on fights it had never seen during training.
This matters because most prediction systems test on data they've already trained on, inflating their accuracy. Our 61% model accuracy and +14.8% ROI on selective bets are honest, out-of-sample numbers from 940 real fights.
See the strategy in action
Learn how the bet signal system turns predictions into profitable selective betting.