Exoplanet Mass Determination Using Transit Data Only: Machine Learning
Abstract
Current methods of determining exoplanet mass require radial velocity observations. However, the majority of the current exoplanet catalog has been initially detected photometrically (e.g., Kepler). This research investigates the presence of natural patterns in photometric transit data using machine learning techniques. A regression based approach uses feed-forward multi-layer neural networks to predict exoplanet mass as a function of known data available from transit observations. Results are compared with true radial velocity based mass measurements, with an average R2 value of -1.142 across a five-fold cross validation technique. Classification schemes are also implemented using the decision tree and k-nearest neighbor algorithms, with comparison of predictive capability between all techniques. For KNN classification, the highest accuracy value obtained was 36.8%. The decision tree classification algorithm was found to be 55.65% accurate. The data set we used was selected from the Exoplanet Orbit Database.
- Publication:
-
American Astronomical Society Meeting Abstracts #233
- Pub Date:
- January 2019
- Bibcode:
- 2019AAS...23346713K