Al Jazeera collected more than 200 metrics measuring the performance history of the 32 teams competing at this year’s World Cup.
These include number of wins, goals scored, FIFA rankings and more, from matches played over the past century, totalling over 100,000 records.
We ran these through Google Cloud’s Vertex AI, which produces state-of-the-art models – adopting Google Brain – that use Efficient Neural Architecture Search (ENAS) to find the best prediction model. The data pipeline is built with GCP-BigQuery.
After every match, the model will be rerun to predict the outcome of the next game all the way through to the final.
Predicting match results isn’t easy. External factors like team morale or player fitness make a big difference to how the game goes.
Based on our simulations for the 2014 and 2018 World Cup tournaments, Kashef is about 71 percent accurate, with an area under the curve (AUC) of some 67 percent.