My latest analysis takes a look at whether a technique called logistic regression can help predict whether a player will score over the next six Gameweeks.
This looks at classifying something as one thing or another (e.g. will score a goal or will not) and assigning a probability to this outcome based on prior data.
Logistic Regression In Action
It turns out that when applied to goal predictions its soothsaying ability is not bad. Using four seasons of data it pushes 50% accuracy.
To help finely tune this I have also looked at the usual suspects that many already use for attempting to predict goals: shots on target, shots in the box, shots, previous goals scored, big chances and penalty area touches both for the whole season as well as last four Gameweeks.
Results
Using this model the top five forwards for the next six Gameweeks are not much of a surprise:
1) Aguero,
2) Kane,
3) Lukaku,
4) Vardy,
5) Costa.
However, with midfielders there are a few surprises, namely a certain Swansea midfielder. This is based on predicted goals not assists by the way.
1) Barkley,
2) Firmino,
3) Andre Ayew,
4) Mahrez,
5) Alli
8 years, 2 months ago
Great stuff. Be interesting to see how well this collection does. Must admit a three fwds line up from the above and five of those mids looks like a great collection for any team.