How does Comète, the Tennis Metrics algorithm, work?

Intro

An algorithm is the description of a sequence of steps allowing us to obtain a result from elements provided as input. For example, the “+” operator is an algorithm for obtaining a sum from 2 or more numbers. If we give it 2, 5 and 7 as input, it will give us 2+5+7=14 as output.

In the context of sports betting, an algorithm is often used to calculate the probability of the outcome of an event such as a player's victory in a professional tennis match. We give it parameters such as the state of confidence of the players, their state of fatigue... And it gives us the output of their probability of victory.

And that's exactly what Comète does!

Well ok, it’s a little more complicated than that… 👇

History and creation

The development of our algorithm began in March 2020. After several months of research and development, Comète was finally born.

Creating a machine learning algorithm is no small feat. Indeed, you must go through 5 key steps:

1. Establish your objective → Finely calculate the probability of victory of a tennis player
2. Find a relevant artificial intelligence technology to answer it → XGBoost
3. Define the parameters that govern its operation → We have precisely 48 of them (example: the ELO ranking by surface)
4. Determine and find the data on which to train → Comète is trained from a database of 20 years of tennis statistics
5. Test its performance on past matches → We tested Comète on the 2018, 2019 and 2020 seasons.

Functionality

Our algorithm uses XGBoost technology, the real movie star of machine learning competitions. Simply put, XGBoost (for eXtreme Gradient Boosting) is an optimized open source implementation of the gradient boosting trees algorithm (trees of what 🤔?!).

Gradient boosting is a supervised learning algorithm whose principle is to combine the results of a set of simpler and weaker models in order to provide a better prediction. We’re also speaking about a model aggregation method. The idea is therefore simple: instead of using a single model, the algorithm will use several which will then be combined to obtain a single result. If you want to know more, we recommend this article.

The idea behind this algorithm is that multiple small algorithms can perform better than a single large algorithm.

To briefly describe the principle, the algorithm works sequentially. This way of doing things will allow the algorithm to improve by capitalisation compared to previous executions. It thus begins by building a first model which is evaluated. From this first evaluation, each individual will then be weighted according to the performance of the prediction, etc…

XGBoost therefore behaves remarkably in machine learning competitions, but not only thanks to its principle of sequential self-improvement. XGBoost indeed includes a large number of hyperparameters that can be modified and tuned for improvement!

These hyperparameters are our secret recipe. Depending on what you choose, it is the performance of the algorithm that is directly impacted. There are billions of possibilities. A fine knowledge of this sport helped us find a combination of 48 hyperparameters among an infinity of possibilities. We can cite a few of them, such as a fatigue parameter, a player form parameter at time calculated from the Elo ranking, or even… In fact, we’re keeping it to ourselves ;)

If you want to have an easy-to-access overview of the different types of machine learning algorithms, we strongly recommend this article on decision tree forests.

Performance

If the probability calculated by Comet deviates too far from the probability reflected by the bookmaker's odds, then a betting opportunity may be on the horizon!

Our algorithm has been in operation since Monday, January 4, 2021 … and I'm sure you're burning with impatience to know its performance!

Here it is: (as of April 19, 2023)

👉 Number of matches analysed: 10,209

👉 Percentage of correct predictions: 59.97%

👉 Number of value bets unearthed: 602

👉 Success percentage: 69.1%

👉 Average value bet odds: 1.54

What is a value bet? Read our article (link to come).

Conclusion

The results of our algorithm are positive despite a difficult 2021 plagued with Covid-19. We made a profit, which already distinguishes us from many competitors (especially if we count only those who are honest 👊).

Nevertheless, 2 downslides ought to be underlined:

• Although very reasonable compared to what is done on the market, our total betting volume is still relatively low to be able to bet exclusively on Comète. Let time do its job.
• Our average rating of 1.54 reflects a rather risk-averse strategy. It reflects a specific betting style that is not necessarily suited to your player profile.

By the way, do you know your player profile ?

This is absolutely essential information to know before betting, and which also changes over time. We have developed a free test tool available by clicking here (coming soon).

Comète must therefore be taken with a grain of salt because it is still young and not necessarily suited to your betting profile. Moreover, it represents 'only' a statistical point of view, a point of view that human beings would be incapable of having as the quantity of data analysed is immense. Consequently, it cannot take into account all the parameters necessary for an exhaustive analysis. It is therefore necessary, but not sufficient. It should be associated with other types of analysis…

At Tennis Metrics, we combine it with 3 other approaches:

• A human analysis compiled by a tennis expert who studies player profiles and match strategy
• A mass indicator that represents the average bets taken by punters around the world
• An odds evolution tracker to understand the reaction of the bookmaker according to the sums wagered

As you will have understood, we are not a tipster in the classic sense of the term. We are a platform that helps you make better betting decisions thanks to indicators that allow you to more accurately measure the real probability of an event's outcome, in order to compare it to that induced by the odds of your bookmaker. .

In short, we help you spot betting opportunities. Better, we help you place high value bets.

For what ? Because it's the only way to win. I'll explain it to you here (link to come) if you're not yet convinced 😉.

Our ONLY way to measure whether we are doing a good job or not is to track the number of subscribers who have improved their results since using Tennis Metrics. And this is what we will soon be able to do thanks to the development of our bankroll management tool…

That's all for today, hope you learned a lot!