Predicting False Positives in Impact Risk

For this challenge, we invite you to become "virtual contributors" to the Asteroid Grand Challenge and develop a hypothetical method, concept note or simple prototype that demonstrates how Machine Learning could be used to help us avoid the same fate as the dinosaurs.

The JPL Sentry system calculates impact probabilities for observed objects. Our idea is to apply a fully connected Neural Network to predict whether a newly observed object will be a a permanent resident on the Sentry Risk Table (due to uncertainty in its orbit), or a "False Positive", where it is put on the Sentry Risk Table and then eventually removed due to new observations narrowing its potential orbit. Since the MPC doesn't keep orbital parameter calculations from the past, instead only listing the orbital calculations that were computed using the most recent observation, the data we really wanted isn't available. If the Minor Planet Center could keep this data archived in the future, that would be a help in this task. So, we make do with using a feature vector of the celestial coordinates of raw observations for bodies from the two tables at JPL's site. In theory the orbital calculations could be recomputed from these raw observations, but we are not familiar with how to do so.

Sentry Risk Table with NEOs that have a non-zero impact risk: http://neo.jpl.nasa.gov/risk/

NEOs which were false positives for impact: http://neo.jpl.nasa.gov/risk/removed.html

MPC DB Search to get raw observations for NEOs: http://www.minorplanetcenter.net/db_search