Transcription
20/10/2011Predictive modeling competitionsmaking data science a sportAnthony GoldbloomCEO, Kagglee-mail oto by mikebaird, www.flickr.com/photos/mikebaird1. Motivation2. Does it Work?3. Why it Works4. How it Works5. Case Studies1
20/10/2011CrowdsourcingMismatch between those with data andthose with the skills to analyse it1. Motivation2. Does it Work?3. Why it Works4. How it Works5. Case Studies2
20/10/2011Tourism Forecasting CompetitionForecast Error(MASE)Existing modelAug 92 weekslaterCompetitionEnd1 monthlaterdunnhumby Shopping Challenge201918% Correctly Predicted Visits171615141312111091234567891011Competition Progress (Weeks)3
20/10/20111. Motivation2. Does it Work?3. Why it Works4. How it Works5. Case Studies4
20/10/2011Kaggle’s Dark Matter Competitionon the White House blog“The world’s brightest physicistshave been working for decades onsolving one of the great unifyingproblems of our universe”“In less than a week, MartinO’Leary, a PhD student inglaciology, outperformedthe state-of-the-art algorithms”User base: 16,000 registered data scientists5
20/10/2011Our User BaseUsers apply different techniques neural networkslogistic regressionsupport vector machinedecision treesensemble methodsadaBoostBayesian networks genetic algorithmsrandom forestMonte Carlo methodsprincipal component analysisKalman filterevolutionary fuzzy modeling6
20/10/2011Not MIT, not SAS UoL?Additional slides7
20/10/20111. Motivation2. Does it Work?3. Why it Works4. How it Works5. Case Studies123UploadSubmitEvaluate &Exchange8
20/10/2011Use the wizard to post a competitionParticipants make their entries9
20/10/2011Competitions are judged based on predictive accuracyCompetition MechanicsCompetitions are judged on objective criteria10
20/10/20111. Motivation2. Does it Work?3. Why it Works4. How it Works5. Case Studies11
20/10/2011Benchmarking12
20/10/2011Untouched problems13
20/10/20112011 3 million prize 25%Outcomes of a competition to predictthe success of grant applications:Successfulgrant applications-Better identify likely successes toavoid wasting resources onhopeless applications-Identify and communicate thecharacteristics of a successfulapplication to future applicants14
20/10/2011Who to hire?Branding: “we do analytics”15
20/10/2011What could the world’s bestanalysts find in your data?e-mail anthony.goldbloom@kaggle.comphone 1 650 283 9781Photo by gidzy, www.flickr.com/photos/gidzy16
Additional slidesNot MIT, not SAS UoL? 20/10/2011 8 1. Motivation 2. Does it Work? 3. Why it Works 4. How it Works 5. Case Studies 1 2 3 Upload Submit Evaluate & Exchange. 20/10/2011 9 Use the wizard to post a competition . Predictive Modeling Competitions Author: Goldbloom Subject:
