SA developers are using a deep-learning engine to figure out who will win in the Rugby World Cup – and their robot thinks the All Blacks will thrash the Springboks
- South African AI and deep-learning developers who have a big office betting pool on the Rugby World Cup started wondering if they can predict the winner.
- They ended up creating Ruggerbot, a prediction engine that uses historical data and a range of factors to figure out which team is likely to win a match – and by how much.
- Ruggerbot thinks the Springboks will win three of its first four matches by big margins, but will be in trouble against New Zealand.
- For more stories go to www.BusinessInsider.co.za.
A South African developed deep-learning system currently thinks the Springboks will do better against Namibia than against Canada, beating the neighbouring country by 63 points while only scoring 55 points against the North American team.
But Ruggerbot, the Rugby World Cup predictor, says the South African team will be beaten by New Zealand's All Blacks in their match on 21 September – by 14 points.
That is not a reflection of their own rampant patriotism, the developers say, and they still have high hopes the Springboks will win the tournament, even though they haven't yet seen the final prediction of their forecasting robot.
That prediction, on the big matches at the end of the World Cup, is still waiting on "a few more data points", developer Thomas Davies told Business Insider South Africa this week, including the weather on the day, which referees control matches, which players are in the starting teams, and what kind of form those players are in.
"The model will constantly be updated, as new data comes in, and we'll be updating the predictions as results come in and players change etc," Davies said.
The software team was inspired by the record size of its office SuperBru betting pool, the group says in RuggerBot's GitHub entry, to see if match outcomes could be accurately predicted. They decided that building a sophisticated model was the only way to be sure.
The team dug up an ESPN rugby statistics database that goes back back to 1896, created a model to rank individual players based on factors they believe impact rugby ability, and then started to include other data that might affect match outcome.
In testing on warmup games the model got the scores wrong, by up to 40 points – but accurately predicted every winner, which convinced the developers they were on to something.
Putting the bot online for sports-lovers and "the fanatical betting person" is mainly an experiment to see how many people notice and how it is used, the team says. But with any luck they'll make some money too.
"We're not really in this for the money, mainly for our SuperBru office pool. Having said that, we'll place a few bets, and see what happens," says Davies.
At the time of publication, this is what Ruggerbot forecast for the Springboks' first four games:
- South Africa vs New Zealand: lose by 11 points
- South Africa vs Namibia: win by 63 points
- South Africa vs Italy: win by 41 points
- South Africa vs Canada: win by 55 points
Receive a daily email with all our latest news: click here.
Also from Business Insider South Africa:
- South Africa has one of the longest workweeks in the world - here's how we compare
- ‘Fraud is a cancer that is crippling our country’: Supreme Court of Appeal upholds 20-year jail sentence for ‘remorseful’ R5 million fraudster
- South Africa’s first 5G network is now live in parts of Johannesburg and Tshwane – here’s what you’ll pay
- We tried the bacon burger for which Burger King gave up its fully-halaal status in South Africa – and we were distinctly underwhelmed
- The South African Police Service has quietly launched a crime-fighting app – which asks for your name and address before you submit tipoffs