Reducing the background in X-ray imaging detectors via machine learning

The sensitivity of astronomical X-ray detectors is limited by the instrumental background. The background is especially important when observing low surface brightness sources that are critical for many of the science cases targeted by future X-ray observatories, including Athena and future US-led flagship or probe-class X-ray missions. Above 2 keV, the background is dominated by signals induced by cosmic rays interacting with the spacecraft and detector. We develop novel machine learning algorithms to identify events in next-generation Xray imaging detectors and to predict the probability that an event is induced by a cosmic ray vs. an astrophysical X-ray photon, enabling enhanced filtering of the cosmic ray-induced background. We find that by learning the typical correlations between the secondary events that arise from a single primary, machine learning algorithms are able to successfully identify cosmic ray-induced background events that are missed by traditional filtering methods employed on current-generation X-ray missions, reducing the unrejected background by as much as 30 per cent.