Today I had the pleasure of presenting at the OptaPro Forum in London alongside some of my favorite analysts, including Simon Gleave and Garry Gelade, and in front of many others: Paul Riley, Mark Taylor, Colin Trainor, Michael Cox, and Omar Chaudhuri, to name just a few. I spoke about creating metrics for players and teams using expected goals – but without basing the models on shots. (Click here for the video.)
The first part of the talk outlined my work to date with shots-based models, summarized here. Next I recapped the “danger zone entries” model posted here in December. And then I offered an extension of the model using five zones, where players received credit for moving the ball between zones with different expected goals. This is a versatile approach that, when combined with assignment of defensive responsibilities, can rate players all over the pitch. The resulting expected goals model was just as predictive of team results as the shots-based model, and it was arguably better at spotting top players.
The final section showed how tracking data could be used to identify more complex situations for an expected goals model. First, I discussed an algorithm that successfully identified 2-on-1 and 3-on-2 breaks. Then I showed how the data could identify pivotal moments even further back in attacking moves – transitions where four or five players sat in front of a midfielder – and explained how passes into an area of maximum options and visibility led to better results. To identify this area, I used the kernels (and k-kernels) of star-shaped polygons. But this was just a demonstration; I think there are plenty of simpler algorithms that clubs can use to rate players based on situations.
It was great to meet representatives of clubs ranging from Chelsea and Brighton to FC Midtjylland and the New England Revolution. I hope they got some food for thought and, in general, a sharper appetite for analytics. In its second edition, the Forum doubled in size in terms of both presenters and audience. I’m sure it will continue to grow.
Click here for a PDF of the presentation.