Losing $3 Million and Being Happy: A Tale of Money, Lives, and Prediction

• Activity Number: 91
• Type: Contributed
• Date/Time: Sunday, August 4, 2013 : 4:00 PM to 5:50 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Bruce Swihart*+ and Ciprian M. Crainiceanu and Brian Caffo and Rafa Irizarry and Yingying Wei and Jeff Goldsmith and Russell Shinohara and Gagan Sidhu
• Companies: Johns Hopkins School of Public Health and The Johns Hopkins University and Johns Hopkins University and JHSPH and JHSPH and Columbia University and Univ of Pennsylvania and University of Alberta
• Keywords: prediction ; machine learning ; Kaggle ; Heritage Health ; competition ; big data

Abstract

The $3 Million Dollar Heritage Health Prize Competition facilitated by Kaggle.com drew over 1500 teams in a period of 2 years attempting to predict the number of days in the hospital of 70,000 insured members given historical claims data. Our department at Johns Hopkins Biostatistics fielded a team and for a year made daily submissions to acquire public leaderboard mean squared error scores for each submission. We reached and maintained a ranking of 8th overall on the public leaderboard. In this presentation we outline some of our methods and administrative tips, detailing the experience of our first Kaggle competition and ultimately losing $3 million dollars and being happy: a tale of money, lives, and prediction.