Reducing False Positives in Kepler Data Through Machine Learning: A Comparative Study of Neural Networks for Exoplanet Detection

With the growing reliance on space telescopes like the James Webb Space Telescope (JWST), which produces up to 57 gigabytes of data per day, applying machine learning for data vetting, filtering, and analysis has become essential. This research focuses on developing and training various types of neural networks using Kepler mission data to compare their accuracy, efficiency, and potential limitations in the search for exoplanets. Specifically, these networks distinguish between exoplanet transit hosts and eclipsing binary systems—a common challenge that causes roughly half of exoplanet candidate false positives in the Kepler pipeline. By improving the ability to differentiate between these signals and analyzing the performance metrics of each method, our research can help significantly reduce false positives, enabling astronomers to prioritize the most promising stars for follow-up observations.