A search of dark matter among Fermi-LAT unidentified sources via the inclusion of systematic features in machine learning

par Mme Viviana Gammaldi (IFT)

lundi 20 juin 2022, 14:00 → 16:00 Europe/Paris

LLR conference room (LLR)

Around one third of the point-like sources in the Fermi-LAT catalogs remain as unidentified sources (UniDs) today. Indeed, these unIDs lack a clear, univocal association with a known astrophysical source. If dark matter (DM) is composed of weakly interacting massive particles (WIMPs), there is the exciting possibility that some of these unIDs may actually be DM sources, emitting gamma rays by WIMPs annihilation. We propose a new approach to solve the binary classification problem of disentangling prospective DM sources from astrophysics among the unIDs of the 4FGL Fermi-LAT catalogue. As a starting point, we work on a parameter space defined by two features, i.e. spectral curvature and energy peak of both the observed (astrophysical) and expected (DM) gamma-ray spectra. We then introduce two synthetic features, i.e. the detection significance and the uncertainty on the spectral curvature. The synthetic features allow us to introduce the systematic uncertainty related to the observation itself into the theoretical data set. The latter leads to an improvement of the overall classification accuracy for all the selected algorithms, namely Logistic Regression, Neural Network (NN), Naive Bayes. We find that the best classifier for this classification problem is the NN. The improvement in the performance of the NN is 8.2% by including the synthetic features, from a 86.2% +/- 0.8% of overall accuracy with two features to a 93.3% +\- 0.7% with four features. We also compare these results with the Gaussian process classification with Noisy Inputs, which return an overall accuracy of 87.0\%\pm 0.1%. Applying the NN to the unIDs sample, we find that the degeneracy between some astrophysical and DM sources can be partially solved within this methodology. Nonetheless, we conclude that there are no DM source candidates among the pool of 4FGL Fermi-LAT unIDs.