HT-ML Predictor

Have you been lucky or not ? This tool can answer your question by estimating the probabilities to win of each team in a Hattrick match already played. The probabilities are calculated with a Machine Learning algorithm, trained with around 3,5 million Hattrick matches.

Match Predictor

Enter your match ID and discover what were the most probable results ! Available for all matches (League, cup, NT and recent tournaments matches)

Enter your match ID :

League Predictor

Simulate all the matchs of your league, for the current season or for previouses. Along the season, who was lucky and who was not ? The calculation time is around 20 seconds.

Enter your league ID :
Enter the season number :

Customizable Match Predictor

Enter the match ratings manually to simulate the match ! Please input real match ratings as they appear in the match report (not hatstats ratings)

Midfield
RightDef
MidDef
LeftDef
RightAtt
MidAtt
LeftAtt
IndSPDef
IndSPAtt
Tactic Type
Tactic Skill

Methodology

This tool is based on a Machine Learning algorithm "Light GBM". It has been trained and tested on a database including the data of more than 2 million matches all around the world (seasons 79-80 - updated on August 29th 2022).

The algorithm estimates the "Expected goals" scored by a team, depending on some features listed below. The "Expected goals" is the average number of goals that should be observed if the match is replayed a lot of times. The probabilities of the results (home win, away win and draw) are deducted from the values of the expected goals, applying a formula close to a standard distribution.

The following features are used for the predictions :
- Final match ratings for the team and its opponent. In case of pull back, the ratings are recalculated to "eliminate" the effect of the pull back.
- Tactic types and skills are taken into account for both teams. The predictions are reliable for teams using long shots tactic too.

Specialties of players following their position on the field have been used in a first version. There are not in the final version, because the predictions are more accurate without this feature. It absolutely doesn't mean that specialities are useless, but their impact on the expected goals is very dependant on the skills of the players, and it is impossible to get this information.

Special thanks to Nickarana for sharing his huge database (used for training the model), and to Jean-Mi for the very useful Python framework pyCHPP !

If you have any question or suggestion about this website, don't hesitate to contact Ludo2293 by HT-mail or in the dedicated post on Hattrick forum.