Analyzing Kepler Data Through Machine Learning: A Comparative Study of Neural Networks for Exoplanet Detection

Classifying exoplanet candidates remains a significant challenge in astronomy, as signals from eclipsing binaries, stellar variability, and instrumental noise often resemble planetary transits. This research builds on the development and training of convolutional neural networks (CNNs) with Kepler mission light curves of confirmed exoplanets and false positives. The current focus is on optimizing the accuracy and robustness of these models and then applying them to Kepler’s catalog of exoplanet candidates. The improved networks generate two outputs: a reduced set of likely false positives and a confidence-sorted list of likely exoplanets, providing a prioritized catalog for follow-up study. By lowering false positive rates and ranking candidate confidence, this work enhances the efficiency of exoplanet validation pipelines and conserves valuable resources by helping ground-based observatories focus follow-up efforts on the most promising targets.