What makes fred tick

FRED is built to analyse football fixtures at scale and turn a large amount of match data into simple, readable predictions.

Behind the scenes, FRED is trained on over 70,000 historical matches from 52 leagues, across 44 countries and 6 continents. The dataset covers more than 5 years of football and is used to train an ensemble of AI models designed to identify patterns in team performance, fixture context and historical results.

Each prediction is built from 168 data points, covering a range of football signals including form, head-to-head records, attack strength, defence strength, team ratings and recent performance trends.

FRED also uses a specialised ratings system to track how teams perform over time. Rather than looking at a fixture in isolation, the system builds a broader picture of each team, allowing it to compare strengths, weaknesses and momentum going into a match.

The models are trained using over 11 million parameters, helping FRED weigh up different signals and decide which factors are most important for each fixture. Instead of relying on one single model, FRED uses an ensemble approach, combining multiple AI models to produce a more balanced prediction.

The aim is not to guarantee results. Football is unpredictable, and that is part of what makes it brilliant. FRED's job is to cut through the noise, highlight useful patterns and give you a clearer starting point for your own research.