bejoselino

The model

It is a Machine Learning model that has been trained on a large number of games (~650,000) to predict the probability of future bets hitting.

What it does

It predicts probabilities of bets hitting for upcoming games. These predicted probabilities are then compared to the odds to find "value bets". If you know a bit about betting principles you will be aware that your only chance of making long-term profit from betting is by consistently finding "value bets" where you have an edge over the odds in offer. Some basic rules I would suggest: bet singles only, don't cash out. The model provides stake size recommendations as a % of your bankroll, these are based off and adjusted version of the Kelly staking criterion.

Performance

The model currently only focuses on "Both Teams to Score" bets. The performance of the model on this market shows a return on investment of 8.6% over nearly 3,000 bets picked. That means every $100 you stake, you would expect to make $8.6. Although this may not seem much to start with, the compounding effect quickly takes the profit into large figures. If you had started with a bankroll of $100, and followed every single one of those bets you would have a bankroll of $76,000 (this in the space of roughly 230 days of betting). Obviously this is the theoretical maximum as it would mean you are able to place bets on all games recommended by the model, whereas sometimes your bookie may not have the game available.

Why I share this?

I started this project as a fun way to learn more about data engineering and machine learning. I am an engineer by profession, and being able to learn this stuff in a football environment seemed awesome to me. I made very good profit following my model for a period of time, but as my bankroll grew substantially the stress from it all took a toll on me and I decided I needed to redirect my efforts into making it more of a service rather than using it for myself. My plan is to share the first few iterations of this model in an open manner.