A new era on Fantasy Surf Predictions
Our new model is up for the season
There has been two years since we’ve started working with statistics for surfing. Like any other project we’ve started small, mining data presented by other websites engaged in the same field. To evolve we felt the urge to build our own database with our own representation of the data and architecture.
All that path led to build solid algorithms for fantasy prediction. Last year our main model has achieved a position near 1kº in a competition with nearly 40kº active players. Our model was grouped around the 2.5% best players of the game at 2017. It clearly shows how well our algorithm has evolved with our deep understanding of how surfer’s variables can be organised to predict the surfers likelihood to do well on an event.
We have struggled a lot, during the season of 2017, to enhance our models. The task has been very difficult and was only achieved by changing our approach.
Now we’ve faced the challenges and started using machine learning techniques on our data to support better predictions. We’re advancing fast on this ground and could already craft a prediction model that is better than all our other models. The benchmark is based on our internal simulations of the 2017 season.
Now that we’re shifting to another field, all our strategies will change. From the data we use to how we process it, it is all diferent now. Starting in a new field is both challenging and revigorating. There are lots of difficulties, but there is also a huge room for improvements.
The first ML model, as we are calling, v2.1 won’t be our main version for the season. We have our internal "version’s championship" and the model will have to be first in our league this year to become official.
Our best model had an accuracy at 2017 around 70%. The WSL fantasy winner of last season, Jase Jase Florence, had an accuracy around 75%. Our ML model, in our internal simulations, had an accuracy around 74%. It got us all, for sure, really excited and thrilled.
Will those models, based on patterns of old seasons, be able to predict well the season of 2018? That's our main question and what keeps us moving!