Single Transit Detection In Kepler With Machine Learning and Onboard Spacecraft Diagnostics

The search for Earth-like planets around Sun-like stars remains at the forefront of the exoplanet field. The Kepler yield loses sensitivity to planets towards longer periods and smaller planets, leaving a dearth of detections within the Earth-like parameter space. To probe longer orbital periods single transit detection algorithms are likely needed. In this talk I will describe a novel technique using an ensemble of Convolutional Neural Networks incorporating the onboard spacecraft diagnostics of Kepler to identify and classify transits within a light curve. We achieve a per-single-transit accuracy of 81% on classifying transits of small-radii (< 3 Earth radius), and long-orbital period (50 - 800 days) exoplanets. We find a significant increase in neural network accuracy when incorporating the onboard spacecraft diagnostics as features. We create a pipeline to recover the location of transits, and the period of the orbiting planet, which maintains sensitivity out to an 800-day orbital period. Our neural network pipeline is set to discover additional planets in the Kepler dataset, and crucially, in the -Earth regime. Additionally, in this talk I will show our first candidate from this pipeline, KOI 1271.02. KOI 1271.02 is the second exoplanet candidate within the system, with KOI 1271.01 experiencing strong (~400-minute) Transit Timing Variations. We perform a series of tests to constrain the orbital configuration and planetary parameters and conclude with a series of potential parameters for KOI 1271.02. Future constraints on the nature of KOI 1271.02 require measuring additional TTVs of KOI 1271.01 or ideally measuring a second transit of KOI 1271.02.