[Paper Review] Hybrid Machine Learning Forecasts for the FIFA Women's World Cup 2019

Andreas Groll, Christophe Ley+3 more|arXiv (Cornell University)|Jun 3, 2019

Sports Analytics and PerformanceEconomics, Econometrics and Finance25 references3 citations

TL;DR

This paper proposes a hybrid random forest model that integrates Poisson-based team ability rankings and bookmaker consensus odds to forecast outcomes of the FIFA Women's World Cup 2019. By training on data from the 2011 and 2015 tournaments, the model simulates the 2019 tournament 100,000 times, predicting the USA as the top favorite with a 28.1% chance to win, followed by France, England, and Germany.

ABSTRACT

In this work, we combine two different ranking methods together with several other predictors in a joint random forest approach for the scores of soccer matches. The first ranking method is based on the bookmaker consensus, the second ranking method estimates adequate ability parameters that reflect the current strength of the teams best. The proposed combined approach is then applied to the data from the two previous FIFA Women's World Cups 2011 and 2015. Finally, based on the resulting estimates, the FIFA Women's World Cup 2019 is simulated repeatedly and winning probabilities are obtained for all teams. The model clearly favors the defending champion USA before the host France.

Motivation & Objective

• To develop a predictive model for women’s international soccer tournaments, addressing the lack of scientific literature in this domain.
• To improve forecasting accuracy by integrating multiple ranking systems into a single machine learning framework.
• To simulate the 2019 FIFA Women’s World Cup 100,000 times to estimate team survival and championship probabilities.
• To compare the model’s predictions with actual bookmaker odds and assess relative performance.

Proposed method

• Train a random forest model on historical match data from the 2011 and 2015 FIFA Women’s World Cups, using team covariates as input features.
• Incorporate team ability parameters estimated via a Poisson-based ranking method that models goal-scoring intensities and maximizes likelihood.
• Integrate bookmaker consensus abilities derived from aggregated odds across 18 online bookmakers using inverse tournament simulation.
• Use the combined set of covariates—including Poisson abilities and bookmaker consensus abilities—within the random forest to predict match scores.
• Simulate the 2019 tournament 100,000 times using the trained model to estimate stage-wise survival and championship probabilities.
• Aggregate simulation results to compute team-specific probabilities of reaching the round of 16, quarterfinals, semifinals, final, and winning the tournament.

Experimental results

Research questions

• RQ1 Can a hybrid machine learning model combining Poisson-based team abilities and bookmaker consensus improve forecasting accuracy for women’s international soccer tournaments?
• RQ2 How do the predicted winning probabilities from the hybrid model compare to those derived from actual bookmaker odds?
• RQ3 Which teams are most likely to progress through each stage of the FIFA Women’s World Cup 2019 according to the model?
• RQ4 Does the model identify structural differences in tournament progression risks between top favorites like the USA and France?

Key findings

• The USA is predicted as the top favorite in the 2019 FIFA Women’s World Cup with a 28.1% chance to win, significantly higher than the bookmakers’ average of 17.7%.
• France has a 14.3% predicted chance to win, slightly higher than the bookmakers’ average of 18.2%, indicating the model sees France as slightly less favored than the market.
• The USA has a 75.5% probability of reaching the quarterfinals, compared to France’s 66.8%, suggesting higher risk of early exit for France despite similar round-of-16 survival odds.
• England and Germany are predicted as the third and fourth favorites, with 13.3% and 12.9% winning probabilities, respectively, both below the USA but above most other teams.
• Lower-ranked teams like Japan (2.7% win probability) and New Zealand (0.3%) have very low chances, with Thailand and Nigeria showing less than 1% chance to win.
• The model’s predictions are consistent with bookmaker odds in general trend but show a stronger preference for the USA as defending champion, highlighting the model’s ability to incorporate team strength beyond market sentiment.