Machine learning competitions Making data science a sport

Machine learning competitions Making data science a sport Jeremy Howard President, Kaggle e-mail jeremy. howard@kaggle. com web www. kaggle. com

Machine learning is everywhere… Insurance companies, banks, operations researchers, hedge funds, mutual funds, derivative analysts, computer scientists, health care organizations, physicians, geologists, astronomers, microbiologists, cosmologists, meteorologists, glaciologists, the federal bureau of investigation, actuaries, search engine companies, utilities, gambling casinos, defense departments, government planners, war games analysts, universities, fraud departments, tourist organizations, economists, sociologists, I-B-M-ers, programmers, politicians, marketing organizations, ad agencies, entertainment companies, physicists, central intelligence analysts, pharmacists, oil companies, mining companies, city planners, retailers, hotels, airlines, movie rental companies, leasing companies, tax planners, climate change researchers, game theorists, credit card companies, endowments, philanthropists, mathematicians, academicians, and philosophers. . .

Crowdsourcing …bit there’s a mismatch between those with data, and those with the skills to analyse it

Global competitions Predicting HIV viral load Competition closes 77% 1½ weeks 70. 8% State of the art 70%

Kaggle’s Dark Matter Competition on the White House blog “The world’s brightest physicists have been working for decades on solving one of the great unifying problems of our universe” “In less than a week, Martin O’Leary, a Ph. D student in glaciology, outperformed the state-of-the-art algorithms”

User base: >32, 000 registered data scientists

Our User Base

Users apply different techniques • • neural networks logistic regression support vector machine decision trees ensemble methods ada. Boost Bayesian networks • • • genetic algorithms random forest Monte Carlo methods principal component analysis Kalman filter evolutionary fuzzy modeling

Tourism Forecasting Competition Forecast Error (MASE) Existing model Aug 9 2 weeks later 1 month later Competition End

Jeremy Howard j@kaggle. com