Part of the winning team contribution to the LSST AGN Data Challenge. See the list of prize winners here.
This notebook explores if classification accuracy of Support Vector Machine and Random Forest algorithms can be improved by removing outliers using an outlier detection method based on robust statistics and manifold hypothesis. The main task of the challenge was to classify stars, galaxies and active galactic nuclei (AGN) using tabular and/or image data.
The AGN Data Challenge is aimed to "motivate planning for AGN science with the Rubin Observatory Legacy Survey of Space and Time (LSST)". The dataset released in this data challenge contains 440,000 astronomical objects pulled from different sources (public archives) and put together to mimic the future LSST data release catalogs as much as possible.
Notebooks describing the full contribution of our team, including the main classification task with supervised and unsupervised machine learning methods, are available on LSST AGN Data Challenge repository.