Spectral Classification of Brown Dwarfs Using Machine Learning

Abstract

Brown dwarfs are compact objects that don't reach temperatures high enough to produce sustained hydrogen fusion. Consequently, they cool over time, gradually evolving through later spectral types. In fact, three new spectral types (L, T, and Y) were added to the Harvard sequence to accommodate the spectral features of brown dwarfs. During the cooling process, some brown dwarfs unexpectedly become bluer instead of redder (at optical and near-infrared wavelengths). This phenomenon, known as the bluing effect, is particularly noticeable at the L/T spectral transition. The aim of this work is to approximate the spectral type of these objects using only photometric data. We used two machine learning algorithms, Random Forest and Gaussian Processes, which were evaluated using a 70/30 train/test split. Both models were trained using 5-fold cross-validation and achieved F1-scores of 0.84 and 0.87, respectively, on the test set. After validating the reliability of the algorithms, we applied them to 21 isolated brown dwarfs without prior spectral type determinations. Our results indicate that 5 of these objects have a spectral type between L0 and L4, while the remaining 16 fall within the M6-M9 range. Machine learning algorithms, combined with multi-band photometry, are a powerful tool for estimating the spectral types of brown dwarfs.