FIGHTintelligence is an independent UFC analytics platform built to understand fights through data, not just predict winners.
The platform combines pre-fight machine-learning probabilities, market odds, fighter analytics, Style DNA and historical performance to provide a structured view of each matchup.
Model probabilities are compared with Pinnacle's margin-adjusted market probabilities to highlight where the model agrees with the market and where it sees the fight differently.
FIGHTintelligence does not claim certainty and is not designed to replace technical judgment. Its purpose is to make UFC analysis more measurable, transparent and explainable.
FIGHTintelligence was created by Vincent Lugat, a Data Intelligence Manager and data scientist with more than 13 years of experience across data science, machine learning, analytics and data products.
The platform is independently designed, developed and operated end to end, from data preparation and feature engineering to model validation, fighter analytics, product design and deployment.
Vincent has followed MMA for more than 15 years and has practiced Brazilian Jiu-Jitsu for several years. FIGHTintelligence grew from the intersection of those two interests: quantitative modelling and technical fight analysis.
Each prediction is generated from a pre-fight snapshot containing only information available before the fight takes place.
Fighter history, recent form, activity, physical attributes, rankings, striking, grappling and matchup-specific features are transformed into a structured set of model inputs.
Historical information is handled causally: statistics from the fight being predicted are never included in its own feature set.
The model outputs a win probability for each fighter rather than a binary pick. This probability is then compared with the market and used throughout the rest of the platform.
A probability is more useful when it has a benchmark.
FIGHTintelligence compares its model probability with Pinnacle's implied market probability after removing the bookmaker margin.
The difference between the two is shown as the edge, expressed in percentage points.
For example, if the model gives a fighter a 60% win probability and the margin-adjusted market probability is 50%, the displayed edge is +10 percentage points.
A positive edge does not mean that a fighter will win. It means that the model assigns a higher probability than the market at that price.
Performance is evaluated using a walk-forward process designed to reproduce real prediction conditions.
The model is trained only on fights available before each evaluation period and is then tested on future fights it has not seen.
As time moves forward, the training set expands and the model is retrained using the newly available historical information.
This avoids evaluating the model using information from the future and provides a more realistic estimate of how it would have performed when predictions were actually made.
The current evaluation process uses monthly expanding-window retraining.
FIGHTintelligence publishes its historical results rather than only highlighting successful predictions.
The Performance section allows results to be explored across different periods, odds snapshots and staking configurations.
Metrics include return on investment, net profit, maximum drawdown, accuracy, Brier score and number of bets.
Opening and closing prices can also be compared to evaluate how results change depending on the market snapshot being used.
Backtest results describe historical performance under a defined set of rules. They are not a guarantee of future returns.
The betting strategy applies to men's fights only. A fight qualifies when the model probability is at least 50%, the odds are 3.50 or lower, and at least one of these holds:
- the model probability exceeds Pinnacle's margin-adjusted market probability by at least 5 percentage points;
- the model probability is at least 70%.
Stakes are expressed in units and scaled by confidence tier: bets meeting both conditions get the largest stake, bets meeting only one get a smaller one.
Model-favored does not mean market favorite. A fighter can have a 51% model probability while still being priced as an underdog by the market. The 50% threshold refers to the model's own assessment, not the bookmaker's.
Style DNA describes how a fighter tends to fight, independently of the model's final win probability.
It converts historical UFC performance into a set of style dimensions covering striking, wrestling, control, defensive ability, finishing tendencies and fight dynamics.
Each score is expressed as a percentile relative to an established UFC reference population. A percentile of P90 means that the fighter scores above approximately 90% of the comparison population on that dimension.
Style DNA is designed to provide context, not to declare that one style is universally better than another. Its purpose is to make stylistic differences between fighters easier to understand and compare.
Archetypes provide a higher-level summary of Style DNA.
Fighters are assigned a primary archetype when their Style DNA profile matches the criteria defined for that style. Secondary styles and modifiers provide additional context when a fighter shows meaningful characteristics outside the primary archetype.
Examples include Grappling Controller, Pressure Wrestler, Submission Hunter, Precision Striker, Power Striker and Volume Pressure Striker.
Mixed Profile is used when no single archetype provides a sufficiently strong representation of the fighter's current profile.
FIGHTintelligence separates current fighter profiles from historical pre-fight data.
Fighter pages use a current profile built from all completed UFC fights available for that fighter.
Historical predictions and completed fight pages preserve the information that was available before that specific fight. They are not retroactively updated with later performances.
For upcoming matchups, Style DNA reflects the fighters' latest available completed-fight profiles.
This distinction allows the site to show how a fighter looks today without rewriting the historical context in which past predictions were made.
The written matchup analysis is generated by a language model from structured fight data: fighter statistics, relative strengths by technical aspect, recent form and matchup context.
The language model does not receive the model's win probability. It does not determine or modify it.
This separation is intentional: the machine-learning model produces the probability, while the language model explains the matchup using the available structured information.
Some fighters have limited UFC history.
When the available sample is small, estimates of form, Style DNA and matchup characteristics are naturally less stable.
FIGHTintelligence identifies these situations with a LOW DATA label rather than presenting all predictions as equally reliable.
The underlying model can still produce a probability, but the surrounding analytics should be interpreted with greater caution.
Current fighter analytics are updated from completed fight data.
Upcoming fights are never treated as completed fights when calculating current fighter statistics or Style DNA.
Dates shown throughout the site distinguish between the date a profile was generated and the latest completed fight included in that profile.
Combat sports are inherently uncertain.
A model can estimate probabilities from historical information, but it cannot observe everything that may influence a fight.
Injuries, training camps, tactical changes, weight cuts, late replacements, personal circumstances and information not present in the dataset may affect the outcome.
Historical statistics can also be noisy, especially for fighters with limited UFC experience or fighters whose style has changed substantially over time.
Model probabilities should therefore be interpreted as estimates, not certainties.
FIGHTintelligence provides probabilistic analysis for informational purposes. Predictions represent estimates, not certainties. Historical performance does not guarantee future results.
Market odds are used as an external benchmark and for historical strategy evaluation. FIGHTintelligence does not accept wagers or operate a gambling service.
If you choose to gamble, do so responsibly, comply with the laws applicable in your jurisdiction, and never wager money you cannot afford to lose.
FIGHTintelligence is independently developed and operated.
Model outputs, Style DNA and written analysis are produced by the platform's own data pipeline and methodology.
Market prices are used as an external benchmark, not as an input designed to force the model toward bookmaker probabilities.