There are a huge number of algorithms and methods for predicting the outcome of a match—the number of goals, fouls, or corners. This article will discuss one of them: determining the outcome probability using the Poisson distribution.
!
What is the Poisson distribution in betting
This is a method from probability theory that, based on the analysis of historical data, allows us to construct the distribution of a random variable—that is, in this interpretation, the value of an indicator in a match—and then derive the estimated probability of a particular outcome. Simply put, we can obtain the probability of an outcome, for example, the probability of "total goals over 2," and convert it into a coefficient, which we can then use in forecasting
The method allows to obtain the probability of the match outcome
Assessing the probability of an outcome is a crucial aspect of forecasting, as a correct probability estimate can lead to a value bet. The Poisson distribution helps with this. In other words, using the distribution is another way, in addition to statistical methods, to estimate the actual probability and the actual odds
A description of the mathematics of the method, which is available on Wikipedia and other websites
There's a classic approach to predicting the number of goals in a match. This involves calculating:
Team attack strength. To do this, take the average number of goals scored by a team in each match and divide it by the average number of goals scored by the home or away team (depending on whether we're counting home or away) across all matches in the league. There's no set number of matches to consider for the calculation; everyone decides for themselves, and each league may have its own optimal value
Team Defense Strength. To do this, take the average number of goals conceded by a team in its matches and divide it by the average number of goals conceded by the home team or away team (depending on whether we're counting home or away team). Again, how many matches to include is an open question.
To get the number of goals for a team predicted by the Poisson distribution, you need to multiply its attack strength by the defense strength of its opponent, and then by the arithmetic mean of its goals in matches (home for visitors, away for hosts)
Attention! The above is a highly simplified description of the method; a more complete one is available on Wikipedia, with three-tiered formulas. In this section, we'll discuss a super-simplified way to calculate this yourself using a calculator.
4SCORE approach
It's worth noting that there's a common misconception that the Poisson distribution isn't a standalone method for determining probability, and there are plenty of calculators online where you just enter everything and get it calculated. So why are we emphasizing this approach? It's one thing to calculate it on your own, and quite another to do so using a huge database on powerful servers, especially with significant algorithmic improvements.
!
We are counting on the “industrial” level
Not just “on a piece of paper with a calculator,” but with our own powerful machines
Statistics database. We use the site's database for the last three years, which is a huge amount of data
Improving algorithms. We improve the classical Poisson using methods of other mathematicians (the Dixon-Koles correction, weight differentiation based on the recency of the match), which allows us to obtain better results
Competitive models. We don't have just one model. We run over 10 models at once with different parameters, and for each league, we select the results of the one that performed best in that particular league
Retraining every day. Our servers work all night to retrain the models on new data for the next game day to ensure more accurate results
Not just goals. The Poisson distribution is actually used for a variety of variables, but many people count goals specifically. Why? After all, a goal isn't random, while a corner is much more so. So why not count corners?
As a result of our machines running, every morning we have a set of outcome probabilities for all matches of the game day. Based on these, we create convenient tools for analyzing events
Poisson-based ML tools
Some might say, "So you got the probabilities of match outcomes, so what?" The good news is that we can now use them in conjunction with the bookmaker's line to search value bets.
1
ML predictions on the website
Using the probabilities generated by our ML machines, we created the “ML Forecasts” tool, which not only highlights the most interesting ones, but also automatically finds suitable odds in the line
We obtain the probabilities. Using machine learning, we predict the outcomes of goals, corners, and fouls for all matches of the day. We do this every morning.
Scanning the line. There we look for those bookmaker odds that are higher than those calculated by ML machines, that is, value bets.
We show this to you directly on the match page or in a separate section for the whole day using the tool ML predictions
2
Poisson calculation for all matches of the day
We provide access to the full results from our ML machines for independent processing (download to Excel) through the tool Match lists. No need to enter everything into the calculator. We've already done that, and everything is ready to download with just one click
Important
This tool does not recommend betting strategies or encourage placing bets. It is a mathematical algorithm designed to highlight interesting predictions for you, but it does not make decisions for you.