Opera Solutions Wins Safe Driver Prediction Competition
JERSEY CITY, NJ, FEBRUARY 26, 2018 — Opera Solutions today announced that Michael Jahrer, vice president of applied machine learning, won first place in Porto Seguro’s Safe Driver Prediction competition, hosted by Kaggle, the premier platform for predictive modelling and analytics competitions. Jahrer’s solution was described by fellow competitors as “beautiful,” “elegant,” “amazing,” and “jaw-dropping” for his use of representation learning to engineer features — a creative job usually left to humans but which proved more effective with a denoising autoencoder.
The competition challenged contestants to build a model that predicts the probability that a driver will initiate an auto insurance claim in the next year. Kaggle reports 5,169 teams entered, with 5,797 data scientists competing. While Michael Jahrer was the sole member of his team for this competition, he often joins his Austrian colleagues for such challenges, all of whom have earned Opera Solutions first-place titles in several global competitions, including the Netflix Prize, the KDD Cup, and the Kaggle Expedia Hotel Search.
Jahrer credits his win to “some basic tricks,” saying “a few things worked out of the box.” He also called this a “classic data mining problem” as opposed to more sophisticated challenges such as those involving images, audio, or EEG data. His solution, however, was anything but classic. He used sophisticated methods typically reserved for deep learning problems, which are the most challenging and involve the most sophisticated datasets such as those used to classify photo, video, and music libraries. He credits his “low creativity” for this decision, when in fact, it was the most creative decision he could have made. No other contestant thought to use an autoencoder to solve this problem.
“Michael’s approach was truly novel and breakthrough — not just within Opera Solutions, but in the machine learning scientific community throughout the world,” said Opera Solutions Founder and CEO Arnab Gupta. “This win is yet another example of Opera Solutions’ ability to conceptualize and leverage new machine learning advances to their fullest and — whatever Michael might say to the contrary — apply deep creativity and extraordinary expertise to the problem-solving process.”
Much of Jahrer’s approach for this competition is mimicked in Signal Hub, Opera Solutions flagship analytics platform. Areas of overlap include the code used for the winning formula, deep learning techniques, and the company’s ensemble approach. Jahrer’s final solution included an ensemble of six models, carefully blended to deliver the final winning score. This approach enabled him to not just edge out the next competing team, but overtake it by a landslide. The difference between first and second place was 0.00285. But in that same numerical distance after second place, a full 25 teams landed. This demonstrates just how tight this competition was and just how much ahead Jahrer’s solution was.
Also leveraging an ensemble approach, Signal Hub is Opera Solutions’ end-to-end Big Data analytics platform for large enterprises. It accelerates the process of extracting insights and intelligence from large volumes of data, including data of different types and in different formats.
Michael Jahrer was a founding member of Commendo, a leading research firm focused on machine learning and the development of mathematical models and algorithms. He’s been with Opera Solutions since the company acquired Commendo in March of 2012, following a virtual tie between Opera Solutions and Commendo in the 2009 Netflix Prize. Download a full report on Jahrer’s Safe Driver Prediction solution.
About Opera Solutions, LLC
Opera Solutions provides software, science, and services that help companies scale analytics across the enterprise and achieve profits and growth that would otherwise not be possible. The company serves global leaders in financial services, healthcare, professional services, the public sector, telecom, retail, travel and leisure, and more through its offices in North America, Europe, and Asia. For more information, visit www.operasolutions.com.