Friday, February 23, 2024
Home > Business Life > World Cup … Why Machine Learning failed

World Cup … Why Machine Learning failed

Machine Learning - World Cup

Just before World Cup, a Machine Learning model appeared with some predications about the path of teams during the subsequent rounds. However, it seems the football itself did not like that. Therefore, it responded by giving us an exciting world cup version that is full of surprises away from predictions.

 Model predications

Machine Learning Model World Cup 2018

  • The model predicted 14 out of the 16 countries reached the second round (round of 16). However, the teams order was not accurate.
  • Model predicated 8 countries to reach the quarter final round. 4 of them left the Mondial before the quarter final (50% accuracy)
  • The model predicted four countries for the semi-final. Only one of them reached. (25%)
  • The model predicated the final match to be between Brazil and Germany. Then, Germany wins! In real world, Germany left the Mondial from the first round. Brazil left from at the quarter final (Zero %). Final match was between France and Croatia, France won.

As an overall judgment, the model failed!

Even when the model predicted 14 out of the 16 countries reached the second round, that was not a significant success. We, humans, predicted the same result without heavy analysis.
Therefore, There is a compelling question:

 Why Machine learning model failed?!

 Short answer, from my perspective, is: It didn’t fail! Problem is: It was miss-used.
Recent technologies (Data science, Machine Learning) are currently used in sports for enhancing the team performance. They even can predict a general trend for the results. However, we should not use that to predict the exact winner of each match (unless that is for academic purposes).

I understand the fact that many models are built focusing on predicting match results mainly for commercial purposes.

Q1. Will they succeed? Answer is they couldn’t till now, But maybe they will make it in the future.
Q2. Is that good? Answer is: No … If technology can precisely predict results, will people follow the match if they already have the results in advance?

Who wins in football?!

If you asked someone “who will definitely win the next match?” the answer is “I don’t know”. We may predict, expect, wish, hope … but nothing is certain because there are multiple factors control the results.

 In football, performance of a team depends on the physical, mental and emotional state of 22 players. Guess what! That mix is a big variable that can change significantly even during the same match. That is why we see big teams lose matches to much weaker teams. Therefore, exact results are out of any calculations.

 Another important factor is Luck. Sometimes, a player may un-intentionally scores a goal. He just wanted to pass the ball to a teammate. However, the ball changed its direction to the sheets. Other times, the ball refuses to enter the goal no matter what plays do.
Obviously, you cannot predict luck.
Actually, that is good news! Part of football beauty is that you cannot predict the match results.

Technology and sports

How data science and machine learning can help sports?
There is a movie called “Moneyball” starring Brad Pitt. The movie tells the true story of “Billy Beane” the American baseball coach. He needed to build a competitive team with a limited budget. He used some sort of data analysis in order to evaluate players.

Moneyball Brad Pitt Analysis

The analysis was based on collecting and analyzing data related to game activities for each player. As a result, He succeeded to build a team with many undervalued players. In 2002, they became the first team in the 100 plus years of American League baseball to win 20 consecutive games. That was a successful example from the previous decade.

Moneyball analysis
A scene from Moneyball

Currently, technology allows collecting huge amount of data during the match and during the training sessions as well. New systems captures live locations of each player, referees and even the ball. Then, they can provide detailed data for player speeds, distance run, pass completion, losing the ball, shots, and much more.
Accordingly, coaches can identify the exact strength and weakness points for each player. They also can evaluate the players’ performance against multiple formations and tactics. Simply, that helps them take the decisions.

Tennis Visualization GIS
Visualizing locations of scored points during a tennis game

There is a direct commercial side as well. Sports clubs and players have fans that follow them on social media. Teams analyze social media fans to identify their segments like geographic locations and age. Then, they can decide locations of new retail stores and the right marketing strategies.

 Same concept is valid for the different sports. In brief, The main target is enhance the performance, not predict the match result.

You may also read:

What you should know about Big Data

Science breaks the records

Football means business