The Model

The model uses the latest techniques in deep learning to predict the outcome of football matches. In order to do so it takes inputs that relate to a team’s historical performances, various rating metrics and bookmakers’ odds.

To train the model we used data from over 50,000 historical matches from all over Europe during the last 20 years. We partitioned the data randomly into training, test and validation sets as the model was tweaked. Finally we ran a 100-fold cross-validation test its performance.

The model has been trained in three different ways:

  • It can output the probability of a home win, away win, or draw.
  • It is also able to output the odds that both teams will score.
  • The model is also capable of generating correct score odds.

From these probabilities we can calculate the fair odds for each game. We can then be compare our odds against a bookmaker’s odds to determine which selections offer a positive expectation, or “value”.

Identifying Tips

During testing we developed a method to generate tips that would be most profitable to users. These tips are consistently profitable across multiple data sets.

Our tips will be accompanied by a suggested stake. You can multiply this stake by as much as you feel comfortable. The suggested stake we provide will never exceed £5.

Test Results

We kept a running total of the profit that we would have made during our cross-validation testing, if we’d followed our own tips. You can see the graph of profit against bets below.


If you’re looking for more data then checkout our data page where you can download some of our cross-validation results and delayed predictions for this season.