Premier League Predictions

Happy Sunday and Community Shield Day to all that celebrate. After watching the unexpected, and rather thrilling match this morning, I could not be more excited for the upcoming season and another year of PremBets.org. Last year was the inagural presentation of this site and methodology, and with the new season, I am looking forward to introducing new prediction models to challenge the KNN and XGBoost.

If you were with us from the start of the Matchweek 6, you would've seen the following changes:

Matchweek 6: Introduced the KNN model to predict over/under 3.5 goals.

Matchweek 8: Included over/under 2.5 goals using a tweaked version of the KNN model.

Matchweek 12: A shift in methodology led to the arrival of XGBoost for over/under with both 2.5 & 3.5 goals.

Champions League Matchweek 6: In early December, I created a moneyline predicting model using KNN and XGBoost for Champions League matches - over/under goals were omitted due to the unpredictability of teams across multiple leagues.

Matchweek 17: The moneyline prediction models were adjusted for Premier League matches

Matchweek 22: While not an actual change, this week observed a perfect 100% accuracy for the KNN over/under 3.5 goals model - Unfortunately, no parlay was placed that week...

Overall, for Premier League, these were the models' performances:

Over/Under 2.5 Goals:

KNN: 49.4% (118/239)

XGBoost: 53.1% (111/209)

Over/Under 3.5 Goals:

KNN: 65.4% (176/269)

XGBoost: 65.1% (136/209)

Moneyline:

KNN: 43.4% (59/159)

XGBoost: 41.5% (66/159)

These results were far from eye catching, however, as the year progressed, the models were producing better results which can be directly attributed to an increase of available data points. With a new season, new players, new coaches, and new teams playing, I anticipate an early struggle for these two legacy models. To combat this, I may introduce data from the 2025-2026 season until Matchweek 8 as this may yield marginal benefits. I am still reviewing the variables of interest and will provide updates on how this impacts the makeup of the models.

And while we are on the topic of changes, here are some anticipated updates to the site:

New Models!!

I would like to utilize LightGBM, CatBoost, and newly released TabPFN-3 - this may require utilizing more historical data from previous seasons. I've used LightGBM for school projects in the past for predicting air temperature in New Hampshire and look forward to structuring the model for something more unpredictable.

New Leagues!!

As a lifelong and eternal fan of Sporting CP, I will be introducing a new page replicating the Premier League predictions for the Portuguese Soccer League - Liga Portugal - to pay homage to my heritage and Grandfather.

New Metrics!!

Although I will still be providing the predictions per models, I am aiming to introduce metrics to average the results across all the models and even set-up an Elo ranking system to show how indiividual teams are currently playing. These changes may not be set-up prior to the start of the season.

More Betting!!

This year will include a page dedicated to 3, 5, 7, and 10 leg parlays for each predicition category based on consensus across the models. Hopefully, this will allow me to recoup some losses from last year's bets and cost of the site...

As always, if you have any recommendations to improve the site or to create a more one-stop shop for Premier Legaue soccer for fans, please reach out to mattclarke6@yahoo.com. Enjoy some Guinness this weekend, and don't forget to support Sporting CP. Visit Sportingcp.ai if you are also an avid supporter. Disclaimer: we are not affiliated with that website.

And thank you for your support with making this website more than just a personal project.