Follow us on


The prediction game

Published : Jul 06, 2002 00:00 IST



BEING a scientist and a football freak need not be mutually exclusive. On the eve of the World Cup tournament at the newly founded School of Applied Medical Sciences and Sports Studies in the University of Ulster, Ireland, a team of sports scientists and statisticians predicted who would lift the cup. The leader of the research team, Peter G. O'Donoghue, is the director of post-graduate studies in sports and exercise at the school.

The scientists used 10 different methods of predicting the results of every match played in the tournament and computer simulation gave the best results. In the simulator-modelling, Brazil emerged the winner in the final. However, the exact prediction was that Brazil would beat Italy in the final. Italy was knocked out by South Korea in the second round.

The methods used ranged from subjective analysis by an individual soccer expert, by a forum of five university soccer-crazy boffins based on pooled knowledge, discussion and consensus and bookmakers' odds to statistical analyses based on regression equations, the probabilistic model using the neural network, analyses of games on a Sony Play Station computer and the use of a simulator that ran the entire tournament 2,000 times over. Of these, the simulator performed the best in terms of predicting the various matches in the run-up to the final.

Brazil won 24.8 per cent of the 2,000 simulated tournaments - the highest - and the researchers gave Brazil 24.8 per cent chance of winning the title. Interestingly, the simulator exercise was the only method that had Brazil in the final. In the predictions using the other methods, France figured five times, Italy three times (therefore, four times including the simulator model), Germany twice, Spain twice, Argentina twice, Mexico, Korea, Nigeria and Japan once each.

The individual expert predicted an Italy-France final, with Italy-Spain and France-Argentina as the semi-finalists; the focus group of five predicted an Italy-Argentina final, with Italy-Portugal and France-Argentina as the semi-finalists; the probabilistic neural network model predicted a Germany-France final, with Germany-Spain and France-Argentina as the semi-finalists; the 'Decision Tree C5.0' model predicted the final and the semi-finals identical to the neural network's; the 'Discriminant Function Analysis combined with a logistic regression analysis' predicted a Spain-Argentina final, with Mexico-Spain and France-Argentina as the semi-finalists; the multiple linear regression-based statistical analysis gave a Korea-Japan final, with Germany-Korea and France-Japan as the semi-finalists; the simulator model gave an Italy-Brazil final, with Italy-Spain and Brazil-Argentina as the semi-finalists; the random scores prediction, based on scores randomly picked from 282 previous matches, gave a Mexico-Nigeria final, with Mexico-Portugal and Nigeria-France as semi-finalists and Nigeria defeating Mexico 6-1; and the Sony Play Station ended up with a Spain-France final and Germany-Spain and France-Denmark as the semi-finalists.

THE researchers gave their home fans little to cheer about. They predicted that England and Ireland would end up runners-up in their respective groups but would lose to France and Spain respectively in the first knock-out stage. Indeed, the result of the Ireland-Spain duel did turn out as predicted. "The result of our research is a classic case of head versus heart - number crunching against subjective judgment," O'Donoghue said.

The key element in the simulator evaluation method was the effect of the following factors on the chance of a goal: the FIFA ranking in June, the distance the teams had to travel to compete, home advantage (with weightage given to factors such as whether the playing nation is the host or not and whether it belongs to the same or different continent), recovery since the last match, the effect of switching between South Korea and Japan and vice-versa for the games and the rest period between matches.

"Italy going out was a major disappointment. Otherwise this would have been the prediction getting both the finalists right," said O'Donoghue. "My only comment is that I am extremely disappointed that the Italian side could not hold on to the 1-0 lead for five minutes against Korea. Italy made the semi-final in 21 per cent of our simulated tournaments, making the final on 15.2 per cent of them and winning the tournament on 8.8 per cent of them," he added in an e-mail response to Frontline.

"Our predictions for Korea were 67 per cent to get through their group, 22 per cent to get to the quarter-final, 3 per cent to get to the semi-final, one in 2,000 chance of making it to the final but no chance of winning the World Cup. They had 0.9 per cent chance of coming third," said O'Donoghue. However, Turkey beat them to the third spot. In the results of the simulation, Turkey lost to Japan in the second round.

In terms of detailed predictions of individual matches, the simulations got five of the eight matches in the last 16 correctly - Germany-Paraguay, Italy-Korea, Spain-Ireland, Brazil-Belgium and Japan-Turkey. England and Denmark were also there but the exercise did not predict that Sweden, Senegal, Mexico and the United States would make it to the second round. To criticism that the predictions had faltered, O'Donoghue said that the critics had the benefit of hindsight. "Not one had shown a more accurate pre-tournament prediction," he said.

He added: "Our model recognised France and Argentina as the two strongest sides, but if one won its group and the other came second they would play each other in Round Two! This severely reduced the chances of either team winning the tournament. Brazil were favoured 86 per cent to reach the quarter-finals as they were to meet someone from Japan's group in the second round. Identifying this sort of thing and the impact on a team's chances was the major contribution of the model."

"Brazil were given odds of 6-1 - we have clearly shown that it should have been more like 4-1 due to the way the tournament was organised. I am satisfied with this, and again would reiterate that I am not aware of any prediction that would have had France, Argentina and Portugal failing to progress from their group. I am aware of a prediction that had Korea in a semi-final; it was based on multiple linear regression and done by us; one of the other 10 predictions - the simulator - is still more accurate than it overall," he maintained.

(This story was published in the print edition of Frontline magazine dated Jul 06, 2002.)



Comments have to be in English, and in full sentences. They cannot be abusive or personal. Please abide to our community guidelines for posting your comment