Predicting the outcomes of football matches is both a science and an art, involving complex data analysis and a bit of intuition. As a first test at predicting Premier League games, I set out to predict the outcomes of matches scheduled between April 20 and April 28, 2024. My goal was to forecast the winners and the number of goals scored by the home and away teams. Here’s a detailed analysis of our predictions and their comparison to the results.
The Process
1. Data Collection:
We gathered historical data for multiple Premier League seasons, including player statistics, match outcomes, and team performance metrics. This data was crucial in training our predictive models.
2. Feature Engineering:
We incorporated recent goal statistics and wage information for each team. The recent goals scored and conceded by teams provided context on their current form. The wage data included the total salary and salaries by position (forwards, midfielders, defensemen, goalkeepers).
3. Model Training:
Using the AutoGluon library, we trained two separate regression models to predict the number of goals for home and away teams. The models were trained using data from all seasons up to the 2023/2024 season, excluding the last 25% of matches from the 2023/2024 season reserved for testing.
4. Predictions:
We made predictions for the matches scheduled between April 20 and April 28, 2024. Our models predicted both the number of goals and the match winners.
Prediction Results
Here’s a summary of our predictions compared to the actual results:
| Match | Predicted Home Goals | Actual Home Goals | Predicted Away Goals | Actual Away Goals | Predicted Winner | Actual Winner | Rounded Predicted Home Goals | Rounded Predicted Away Goals | Correct Winner |
|---|---|---|---|---|---|---|---|---|---|
| Luton Town vs Brentford | 1.158365 | 1 | 2.310192 | 5 | Away | Away | 1 | 2 | 1 |
| Sheffield United vs Burnley | 1.049577 | 1 | 1.687838 | 4 | Away | Away | 1 | 2 | 1 |
| Wolverhampton Wanderers vs Arsenal | 0.458599 | 0 | 2.693839 | 2 | Away | Away | 0 | 3 | 1 |
| Everton vs Nottingham Forest | 1.583769 | 1 | 0.90472 | 0 | Home | Home | 2 | 1 | 1 |
| Aston Villa vs AFC Bournemouth | 2.570876 | 3 | 1.108545 | 1 | Home | Home | 3 | 1 | 1 |
| Crystal Palace vs West Ham United | 2.520274 | 5 | 1.38622 | 2 | Home | Home | 3 | 1 | 1 |
| Fulham vs Liverpool | 1.250518 | 1 | 2.014094 | 3 | Away | Away | 1 | 2 | 1 |
| Arsenal vs Chelsea | 2.897227 | 5 | 1.143637 | 0 | Home | Home | 3 | 1 | 1 |
| Wolverhampton Wanderers vs AFC Bournemouth | 0.631473 | 0 | 1.405626 | 1 | Away | Away | 1 | 1 | 1 |
| Crystal Palace vs Newcastle United | 2.641256 | 2 | 0.993894 | 0 | Home | Home | 3 | 1 | 1 |
| Everton vs Liverpool | 1.067201 | 1 | 1.898021 | 0 | Home | Home | 1 | 2 | 1 |
| Manchester United vs Sheffield United | 2.562534 | 4 | 0.845323 | 2 | Home | Home | 3 | 1 | 1 |
| Brighton & Hove Albion vs Manchester City | 0.71447 | 0 | 2.385597 | 4 | Away | Away | 1 | 2 | 1 |
| West Ham United vs Liverpool | 0.985477 | 2 | 1.80883 | 2 | Away | Draw | 1 | 2 | 0 |
| Fulham vs Crystal Palace | 1.697959 | 1 | 0.862809 | 1 | Home | Draw | 2 | 1 | 0 |
| Manchester United vs Burnley | 1.715171 | 1 | 1.511557 | 1 | Home | Draw | 2 | 2 | 0 |
| Newcastle United vs Sheffield United | 3.657062 | 5 | 0.803784 | 1 | Home | Home | 4 | 1 | 1 |
| Wolverhampton Wanderers vs Luton Town | 0.977961 | 2 | 1.294268 | 1 | Away | Home | 1 | 1 | 0 |
| Everton vs Brentford | 1.162282 | 1 | 1.752335 | 0 | Away | Home | 1 | 2 | 0 |
| Aston Villa vs Chelsea | 1.924533 | 2 | 1.228519 | 2 | Home | Draw | 2 | 1 | 0 |
| Accuracy | 40.00% | 30.00% | 70.00% |
Prediction Accuracy
The accuracy of our predictions was as follows:
• Winner Prediction Accuracy: 70.00%
• Home Goals Prediction Accuracy: 40.00%
• Away Goals Prediction Accuracy: 30.00%
Analysis and Insights
• The model performed well in predicting the match winners, with an accuracy of 70%. This indicates a solid understanding of overall team strengths and outcomes.
• Predicting the exact number of goals proved to be more challenging, with the model achieving 40% accuracy for home goals and 30% for away goals. This highlights the inherent variability and unpredictability in soccer matches.
Conclusion
Our machine learning model showed promise in predicting match outcomes, particularly in determining the winners. While there is room for improvement in predicting the exact number of goals, the results are encouraging and provide a foundation for further refinement and enhancement of the model.
As we continue to gather more data and refine our models, we expect to see even better performance and more accurate predictions in the future.
Feel free to share your thoughts and insights on our prediction model. How do you think we can improve our predictions further? Let us know in the comments below!
