Methodology

Overview

BoutMetrics predictions are generated by an ensemble of machine learning models trained on 8,500+ historical UFC fights dating back to 1994. Our system processes fighter statistics, historical performance, stylistic matchups, and real-time odds data to produce win probabilities, method predictions, and value-betting signals.

Three Core Models

• Decision Model — Predicts whether a fight goes to the judges' scorecards and evaluates over/under round totals.
• KO Model — Estimates the probability of a knockout or TKO finish, factoring in striking power, chin durability, and pace.
• Survival Model — Analyzes inside-the-distance finishes and round-by-round survival probabilities using time-to-event modeling.

Validation

All models are validated using a strict walk-forward methodology: training only on data available before each event, then predicting unseen fights. We never use future information or cherry-pick results. Our published accuracy of 60.9% and ROI figures are based entirely on these out-of-sample walk-forward tests against real sportsbook closing lines.

Data Sources

Fighter statistics are sourced from official UFC data, supplemented by historical fight records. Odds data is collected from multiple licensed sportsbooks and prediction markets in real time. We do not use proprietary or insider information.

Limitations

No prediction model is perfect. MMA is an inherently volatile sport where upsets happen regularly. Our models cannot account for undisclosed injuries, weight-cut issues, or fighter motivation. Predictions are probabilistic estimates, not certainties. Always exercise your own judgment and gamble responsibly.