Recent Updates and Key Insights

Happy Sunday and end of January! I have been looking to utilize the blog portion of this page more as a way to express my thoughts on this page in relation to the models, methodologies, results, and new features, as well as providing some personal commentary on the Premier League matches. Above all, I am a fanatic for soccer in all forms - a lifelong Sporting CP and Portuguese soccer supporter. Last season was the first time that I would consider myself an avid viewer of English soccer, watching any games that were being broacasted on USA Today (I refuse to pay for a Peacock subscription). With the move to bring in Ruben Amorim to Manchester United and the brilliance of Bruno Fernandes, naturally I have found myself to claim Manchester United as my "team". Of course this season has looked substantially better for United, I am saddened by the sacking of Ruben...

This season has been full of spectacular matches and unexpected outcomes such as Sunderland immediate Premier League success, Aston Villa challenging Aresenal and Manchester City for the title race, and Tottenham struggling once again - potentially in relegation territory. And with the recent transfers between clubs, the changing landscape has brought more excitement for the second half of the season. Similarly to more American based leagues, it is unique to see the countelss inter-league transfers that can be seen as helping an opponent ie Isak moving to Liverpool from Newcastle United, Semenyo moving to Manchester City from AFC Bournemouth, Garnacho moving to Chelsea from Manchester United, and other moves that were not as publicized. While the transfer fees seem to be ever increasing, the never-ending rumors and purusit for money makes it tough for the average fan

to know who will be playing in any given week. Coming from a Sporting CP fan that lost Gyokeres to Arsenal and will be waving goodbye to Quenda as he departs for Chelsea in the Summer - hopefull after helping Sporting win the league for the third year in a row. As my grandfather says, "It's all about the money, and I never see them use it to improve the team", which is less apparent in a high quality league like the EPL. I think that brings out the excitement and support for teams like Sunderland and Aston Villa, while creating more animosity towards clubs like Manchester City and Liverpool. And I am no different, except for Gyokeres' move to Arsenal since his performance for Sporting CP the past few seasons were electric.

As I bet less and watch more, it has highlighted the variability in games and matchweeks while also highlighting the struggles of predicting sports. Yes, you can start to see trends in data and identify likely outcomes, the cyclicality of performances and unexpected impacts to the game cannot be predicted. While you can likely expect Canceido to receive a yellow card, it's harder to predict Richarlson's hamstring injury early in the first half or Wolves to win their first game of the season against Manchester United. This has led me back to the roots of this project, to understand and improve accuracy through different methodologies rather than making money. Although, the Over/Under 3.5 KNN model was 100% accurate for the first time all season in MW 22.

It is exceedingly ill-advised to immediately bet money on predictions made by a black-box algorithm without fully observing it's limitations and exploring it's capabilities. A prime example would be with the Moneyline predictions - currently the KNN model is leading with ~44% accuracy, which is abysmal, but the XGBoost 3.5 Over/Under model is nearly 70%. A quick generalization would suggest that the talent and competitiveness in the Premier League causes these models to have less success, but as the curator of these models, I would argue that the variables being fed into the model are not optimally selected and are too similar to the Over/Under models' inputs. These insights are easy to reflect on, but are more difficult to act on, since realistically, each week

the models should naturally be improving due to more historical data being available for training the models. Due to starting a new job and acclimating to a new city, I have had limited free-time to improve and refine the models, and find myself struggling to maintain the records page and running the model for the next matchweek. In my last post

I outlined multiple improvements that I want to explore, but in reality, I will be prioritizing creating a pipeline that automates the weekly running of the models as that will allow more time for other improvements to the code and site.

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. Stay warm - I hope you all watched the Sporting CP masterclass aganst PSG and Benfica shocking the world to keep their Champions Leauge endeavors alive. Visit Sportingcp.ai if you are also an avid supporter. Disclaimer: we are not affiliated with that website.