Out of all spheres of technological advancement, artificial intelligence has always attracted the most attention from the general public. Apart from actual developments in the field of AI, the massive body of sci-fi literature and movies is partially responsible for this fascination. The concept of True AI - a hypothetical reality in which AI has the same learning and intellectual capacity as humans - is truly astonishing.
However, in the real world, we’re still pretty far from True AI. That doesn’t mean we’re not already heavily reliant on AI in our everyday lives, though. Machine learning, a subset of artificial intelligence, is behind the biggest leaps in the field.
In layman’s terms, machine learning is a study of algorithms that analyze huge sets of data in search of structure and patterns. Based on this data, machine learning algorithms can ‘learn’ by themselves, improving independently over time, without additional programming. Furthermore, they’re able to provide predictions on different outcomes thanks to the data they’ve analyzed.
As we’ve mentioned, we’re already dependent on AI and machine learning in many industries. Most suggestion algorithms and content detection in services like YouTube and Netflix use machine learning. Search engines, image detection, speech recognition, financial analysis - the list goes on.
One sphere in particular is witnessing an increased use of machine learning: predicting sports matches. And that is what we will discuss in this post.
The majority of machine learning appliances perform tasks that human actors used to perform. Predicting sports results is no different. Coaches, sports experts, and of course bookmakers have been creating their own match-result predictions since the birth of professional sports.
Bookmakers and betting markets in general are huge drivers for the advancement of machine learning use in predicting sports outcomes. The incentive is more than obvious - the sports betting industry was valued at $85 billion in 2019 .
Many prediction markets are already surprisingly accurate . Election predictions are often on point, but even in some cases when the general consensus among analysts was wrong, as was the case with the 2016 US presidential election, betting markets were right .
In general, sports betting markets - especially for soccer - tend to be very precise. Why the need for machine learning, then? Well, one of the main reasons is the increased precision of predictions. Machine learning algorithms can scour huge sets of data and extrapolate outputs (predictions) at a scope and pace that’s hard for humans to match.
Machine learning algorithms are fed huge amounts of data from which they learn and create predictions. The data used is basically the same that humans use to anticipate match outcomes. In soccer, for example, this includes team and player performances like the number of goals, passes, and possessions for each player, as well as the results of previous matchups between the teams.
Machine learning for sports predictions largely relies on building a classification model based on a training data set; the initial data is fed to the algorithm so it can detect patterns and create predictions. There are two training methods: supervised and unsupervised. The former builds prediction models based on both input and output data, while the latter only includes input data.
The trickiest part of predicting match outcomes with ML is choosing the right data set. Machine learning predictions aren’t as simple as throwing in all the data you believe to be relevant and expecting instant results. The predictions are of varying precision, and are often still less precise than those made by bookmakers .
That’s why some research uses bookmaker odds as an additional factor to increase precision in machine learning algorithms. The use of artificial neural networks (ANN) - which are modeled after the human brain - allows for the addition of even more factors. This in turn leads to even more precise predictions, but their accuracy is still generally between 50-70% .
Of course, no one believes that sport outcome predictions will reach 100% accuracy, or even get near it. This is impossible due to the inherent randomness of events, the same randomness that makes betting a beloved pastime for many.
The question that has probably crossed everyone’s mind is: Why don’t we use machine learning to correctly predict sports outcomes and become rich through betting? Well, there are several reasons.
The first is that bookmakers are already ahead of the game . If you believe that odd makers are oblivious to the advancements in machine learning, you’re dead wrong. They have their own data analysts who help them improve their predictions and adjust the odds accordingly. No matter how advanced machine learning gets, you can count on bookmakers to be several steps ahead of the average punter.
The second reason is that the uncertainty of predictions, even those of bookmakers, means that you’d need a huge amount of time - and theoretically infinite capital - in order to see returns in the long run. Furthermore, the more factors you add into your machine learning algorithm, the longer the training period for the algorithm is, further prolonging the time before you see potential returns.
Machine learning isn’t hidden, esoteric knowledge - all the relevant actors are very much aware of its capabilities and use it to its full extent. If you have any get-rich-quick ideas involving machine learning, we suggest you ditch them.
Lastly, even if you theoretically managed to get a surprisingly high precision rate, bookmakers would probably start discriminating against you as soon as they saw a huge win rate over an extended period of time.
While beating the betting market might be impossible, predicting sports match outcomes has plenty of uses. Besides being used by bookmakers to better determine the odds, machine learning helps a whole string of people employed in the sports industry. Coaches, sports analysts, and managers rely on it to anticipate their future performances and analyze previous ones. Many regular users use machine learning to simply improve their data skills.
Beyond predicting outcomes, machine learning has found plenty of other uses in the world of sports. Individual player development, predicting possible injuries, classifying players - the uses are plentiful . While most of us probably won’t get rich off of machine learning, the sports industry - and the world in general - are better because of it.
This Guest Post is written by Ilija Acimovic who is a sports correspondentcomments powered by Disqus