A machine learning side project for ice hockey analytics.
In 2023, I challenged myself to apply machine learning to a dataset near to my heart: predicting NHL hockey game outcomes. No, I won’t be retiring off the back of a successful NHL betting career any time soon. But I did achieve my goal of building a model that could hang with the big boys, performing in line with other public models.
Sports prediction is hard. The best models I’ve seen for the NHL are around 55% accurate in their predictions. That makes it a great challenge for applied machine learning! Aside from making an interesting portfolio piece, this project was useful to sharpen a number of skills:
- API data ingestion via the NHL’s “unofficial” API
- Data science workflows with the excellent tidymodels framework
- Putting the model to the test and summarising the results in a data story with Quarto
View the slides here or interact with the embedded presentation below.