NMPi Predicts The World Cup Using Machine Learning: Man vs. Octopus vs. Machine

It was a trip to the grocery store one Sunday evening in May that I first saw it, The World Cup 2018 Panini Sticker Book. Unmatched in its dual role of footballing holy grail and financial black hole. With a quick glance over the shoulder to make sure no one was looking, I placed it delicately into my basket. After all, such a guilty pleasure required some degree of secrecy to ensure the maintenance of dignity.

Since then it’s all got a little out of hand really. My flatmates and I are fifty stickers away from completing the book – 650 in total. The team at the local Co-op know us by name – I’m now greeted with, ‘the usual, Alex?’, as I walk in.

To the utter bewilderment of those around me, the World Cup and its advertising machine has gone fishing and managed to catch a whale. I am that whale. But rather than let that get to me, I’ve decided to apply my enthusiasm in a slightly more useful way.

About a year ago we wrote a blog article predicting the results of the UK election, correctly arriving at the verdict of a hung parliament. So, we have decided to revive our predictive capabilities for the World Cup.

Fred Maude, the genius behind our previous efforts, has been busy hoarding reams of player data to apply to his new footballing-based algorithm.

We’ve incorporated machine learning into our business in a variety of different ways (award-winning in some cases), and Fred has been at the forefront of that movement. It works for our Google Shopping campaigns. It worked in 2017’s UK election. Will it work for Russia 2018? A few words from him later, along with his results.

But first, we thought we’d up the stakes a little. Alongside Fred’s machine, we’ve asked two very different figures to put their theories forward. After all, who wants machines to win every time? We all know how that ends…

Man – Max Flajsner, Head of Performance

Ignoring the fact Max is a Spurs fan, and so has little hope of being able to identify the characteristics of a successful team, we thought we’d let him have a go anyway. With a solid footballing background and a leading role in my sticker book efforts, he is perfectly placed to represent all of humanity.

In explaining his approach, Max revealed “my strategy mainly rested on Mousa Dembele. I just love him. So, to hear he won’t be starting had a huge impact on my decision-making. Belgium out in the quarters, Germany to ruthlessly defend their title.” Bless him.

Image Source: Sky Sports

2nd Round – Colombia v England, Uruguay v Portugal, France v Croatia, Brazil v Mexico, Belgium v Poland, Spain v Egypt, Argentina v Peru, Germany v Costa Rica

Q’ Finals – England v Germany, Portugal v France, Brazil v Belgium, Spain v Argentina

S’ Finals – Germany v Spain, Brazil v France

Final – Germany v Brazil

Winners – Germany

Octopus – Unnamed, Long Suffering Girlfriend

8 years ago, Paul the Octopus shot to fame by successfully predicting the results of the 2010 World Cup. Unfortunately, one thing I’ve learnt in this endeavour is that there are very few publicly accessible predictive octopi left.

So I turned to my cat (Pablo), perhaps the most single-minded creature I have ever come across, which I respect immensely. Whatever we tried, he did not want to participate in our ‘pick a flag with some tuna on it’ game. As he sat there staring at me, his eyes questioning how on earth it had come to this, I knew I had to find another way.

My final option was to go for a creature who matched Paul & Pablo in sporting awareness, though perhaps not in the number of legs. She has two legs, cares very little for football, is awesome, and just so happens to be my girlfriend. Upon being asked about her predictive strategy for this year’s tournament, she replied, “if I’m completely honest Alex, I just want this all to be over.” Perfect.

After some negotiations around where I would buy dinner, I managed to eek out an infographic. My goodness, I hope she’s right.

Image Source: Sky Sports

 2nd Round – Colombia v England, Uruguay v Portugal, France v Argentina, Brazil v South Korea, Belgium v Japan, Spain v Russia, Nigeria v Denmark, Germany v Costa Rica

Q’Finals – Germany v England, Portugal v Argentina, Brazil v Belgium, Spain v Nigeria

S’Finals – England v Spain, Portugal v Brazil

Final – Spain v Portugal

Winners – Spain

Machine – Fred Maude, Performance Manager

Finally, the one we’ve all been waiting for. Get your betting apps at the ready, because this one’s the winner. Here’s Fred…

‘The algorithm was designed to account for individual player performance over the last 12 months, along with national team form and results. This included a penalty bot should a knock-out game end in a draw.’

With Croatia and Morocco in the semi-final, and Argentina not making it out of their group, Fred’s machine has put its neck on the line. However, if he is right, we’re in for one hell of a tournament.

Image Source: Sky Sports

2nd Round – Senegal v England, Russia v Morocco, France v Nigeria, Brazil v Sweden, Belgium v Poland, Spain v Uruguay, Croatia v Australia, Germany v Costa Rica

Q’Finals – Germany v Senegal, Morocco v Nigeria, Brazil v Belgium, Spain v Croatia

S’Finals – Germany v Croatia, Brazil v Morocco

Final – Germany vs Brazil

Winner – Brazil

“I’m confident my predictions won’t be far off and my methodology is sound.’

“Initially, the aim was to develop a technique that was devoid of statistical and human bias. We could have gone away and taken stats upon stats to thread into our Machine Learning algorithm, however, we would essentially be feeding the machine with stats that hold no relevance for the game in hand. As everyone knows, it is not beyond Man City to be beaten by Stoke, although player statistics would not point you there. Thus, to remove any bias, and allow the machine itself to learn and make assumptions and stats for itself, we chose a different method.’

“This method involved taking the team line ups and results from over 5000 games in the last 12 months and then feeding this data into our classifier. My first experiments involved using the standard decision tree classifier, however, it was clear that the results were marred with inconsistencies. The decision to drop this and adopt the random forest technique was recently validated by an MIT Machine Learning algorithm aimed at the same outcome as ours.’

“To test the success rate, we looked at previous games with known results and supplied the algorithm with only the team’s line up data. Using this data, the machine then made a decision based on what it thought the result would be. We were then able to match these against the real results. My tests logged a success rate of 64%. Considering the volatility of football, as I am sure anyone who watches the game understands, we considered this as a sign of success.’

“Now that everything had been tested to our satisfaction we could finally predict the winners of The World Cup 2018. The only input data for prediction being the expected line ups of each World Cup team. Unfortunately for those die-hard England fans, a game against Senegal in the round of 16 may prove all too much.”

 

And there you have it. Tune in in 6 weeks’ time to see who triumphed in what Netflix is already touting as its next feature film – Man vs. Octopus vs. Machine.