The Heritage Health Prize is one of the best-known competitions on Kaggle. Yesterday, at an Institute of Analytics Professionals of Australia seminar on healthcare analytics, Phil Brierley – one of the members from the team that won the last round – described how they did it.
Phil is well-known as a neural network guy, but that’s not what his talk focussed on. The most interesting takeaway for me was the process they went through:
- They contacted others on the leaderboard.
- They negotiated how they’d split the prize money if they teamed up.
- They formed team POWERDOT (a name comprised their initials).
- They averaged their algorithms into an ensemble.
So there you go, it seems data science really does benefit from social and soft skills. 😉
Unfortunately there’s no recording or slide deck available from the talk, but the speaker claims most of the technical information is available on this post from a year ago.
Teamwork and ensemble methods have a great winning streak when it comes to predictive analytics competitions – the Netflix prize was won that way as well. Something else both stories have in common: neither solution made it into production. That’s worth pondering.