Predicting earthquakes could be life saving, but the connection between certain types of electromagnetic signals and earthquakes is still debated.

NASA and the Quakefinder team wanted to develop new algorithms capable of searching terabytes of data and identifying specific types of electromagnetic signals that may precede earthquakes by days or even weeks. The algorithm would need to successfully learn how to distinguish between false-positive signals (from sources like lightning) and those originating from Earth’s crust.

With an Analytics Starter Pack from Topcoder, NASA and Quakefinder received on-demand access to more than 60 data scientists who worked to develop new algorithmic approaches to the problem. Competitors were provided with 3TB of data, including known/identified quakes to help "train" algorithms, as well as data sets with "hidden" quakes to test/validate algorithms.

Topcoder developed more than 100 possible algorithmic solutions within three weeks, and the top six solutions each identified up to seven earthquakes with statistically significant scores.

Ultimately, Quakefinder hopes to incorporate the best features of these and other external algorithms now in development. These combined algorithms will become part of their daily search for earthquakes using data from 165 instrument sites in California and international locations.

61 Data Scientists
61
Submissions
101
Winning Solutions
6
Duration
3 Weeks