Asteroid Classification with GAIA: Machine Learning Insights from Reflectance Spectra

Asteroids are diverse objects found in the Solar System, populating in large numbers the near-Earth space, yet most of them form the main belt. While upcoming observatories promise frequent observations of asteroids to compute their orbit, understanding their physical composition remains a challenge due to the inherent difficulty in obtaining spectra.

Launched in 2013, GAIA satellite repeatedly scans the entire sky, not only mapping the more than 1 billion stars, but also discover and measure exoplanets, asteroids, comets and planetary satellites in our Solar System. For the asteroids and comets, GAIA provides more accurate astrometry, photometry as well as albedo and spectra.

In this study we accessed the freely available GAIA Solar System Objects reflectance spectra in the visible range (418-990 nm) and built a catalogue of their spectral wavelengths. After thresholding for the well calibrated spectras we tested 8 unsupervised machine learning (ML) models to cluster the spectras into classes.

After fine hyperparameter tuning of the models we found that several ML methods capture well the features of the already classified asteroids. Also, we obtained a spectral classification for the unclassified asteroids observed by GAIA.

The next step is to extend the study by employing the dynamical features and the albedo to better constrain the asteroid families, and ultimately, to understand distribution of the asteroids in the Solar System.

Acknowledgements: SA was supported during this work by the ESA Archival Research Visitor Programme.