Machine Learning and the future of betting
Congratulations!!! You’ve just won the Nigerian Lottery. All you need to do is send $5000 to us to release the funds…
There’s a fair chance that you’ve seen a similar headline somewhere in your spam folder. However what’s interesting here is how that email made its way into your spam folder. To filter out what is considered spam, whilst still inboxing legitimate emails, Google uses machine learning.
We’ve already looked at betting systems and syndicates and the ways in which we can use data to get an edge in our betting. Like stock markets, betting markets are essentially just predictions of a certain outcome. And to frame a market (or a stock), we need to incorporate a range of factors and construct a prediction of what that outcome might be.
Machine Learning and Artificial Intelligence are certainly buzz words at the moment, however they are effectively just another subset of data analysis that can be used to help predict outcomes.
What is Machine Learning?
By Wikipedia’s definition it gives “computers the ability to learn without being explicitly programmed”.
So for horse racing that might be the fastest sectional in a trial. It makes sense that there is a fundamental relationship between the factor and the outcome of a race.
From that point we go on to test that factor against our database to determine if there is in fact an edge that we can use to make a profit.
Machine learning on the other hand takes this process one step further. Instead of creating an algorithm or a system based on factors we choose, machine learning effectively seeks to find those patterns itself. It is learning from the data that we provide.
Our spam filter is a great example of how machine learning works. If you have a number of emails that you consider spam, you’re able to effectively train the algorithm to detect whether or not a future email contains all the markers of a Nigerian scammer.
The algorithm will assess spam emails and search for keywords that are prevalent in the data and then apply what it has learnt to do its best to assign a probability to future emails that they are either spam, or they’re from a legitimate source.
The application of machine learning to sports betting and horse racing appears to be obvious.
Ideally the machine learning algorithm would be able to identify factors that lead to a win. It may also look at certain factors that are negative for a horse’s winning chance. It can apply those learnings to future events. As time goes on the training set gets larger and the algo continues to evolve and it effectively gets smarter.
Take AFL for example.
The algo would be able to look at all the factors that we have available to us. Wins, scoring shots, player rankings, time in forward half, pressure acts, disposal efficiency etc. Then it would identify which factors are important and weight them accordingly. The algo trains itself and then constantly tests and improves. For future matches it simply applies what it has learnt to give a probability of a team winning. A big caveat here is that both the game and the betting market are constantly evolving.
One very switched-on punter had this to say about machine learning:
‘I use machine learning techniques to analyse data related to sports betting.
It allows me to take a more ‘macro’ approach to betting as the models can be constructed over a wide variety of sports which makes things more scalable and can increase turnover. There is a lot more value in machine learning rather than other basic statistical techniques as it does not confine itself to linear analysis like regression does. It can spot patterns and trends in data that are not necessarily easy to spot on the surface.
I believe in the future that most bookmakers and more punters will start using more sophisticated technology like machine learning which will start to erode the edge. Also as computing power becomes stronger and stronger the greater edge may be left with a select few who have the resources to fund the best technology.
In the interim there are a lot of software packages and online communities out there that can help the novice get up and running in building machine learning models. I would highly recommend this link to someone who was interested in trying out the model building process.’
There are already a number of large betting syndicates worldwide that are at the forefront of trying to use machine learning to predict outcomes of sports as well as hedge funds and proprietary trading firms.
We are also seeing machine learning applied to other things like trying to predict the outcome of elections, the stock market, fraud detection and marketing.
If you’ve ever scanned through your spam filter, you’ve probably seen that quite a few emails from legitimate sources still end up there. Similarly, we’ll occasionally get spam in our inbox.
This goes to show us that anecdotally, the process is still far from perfect.
Like our other betting systems and any quantitative strategy used for investment, one of our biggest challenges is overfitting. Also known as backfitting, it occurs when we start to find relationships in our data set that may or may not be related to the outcome of an event.
Overfitting is the clear enemy of machine learning. The whole process of machine learning is about finding relationships, which means there is always a chance that an algo will overstep the mark. An occasional spam email getting through isn’t so much of a concern in our daily lives, however when there’s money on the line that becomes much more important.
Of course that’s why we need to ensure that we are always using multiple datasets to firstly build models and then test them.
However in years to come there is no doubt that things like machine learning, artificial intelligence and quantum computing are going to play an increasingly prominent role in not only our betting but in our everyday lives.