Training a convolutional neural network for exoplanet classification with transit photometry data.

The search for exoplanets aims to identify planets with compositions similar to Earth's, providing insights into planetary formation and habitability. As a result, efforts to enhance the efficiency of exoplanet research have led to the development of various detection methods, including transit photometry. Despite their effectiveness, these methods produce data that require detailed interpretation, such as identifying dips in light curves. Machine learning has then emerged as a powerful alternative, offering rapid image classification and the ability to analyze complex datasets in a short span of time. This paper applies a convolutional neural network (CNN) to the Kepler dataset, which consists of time-series light curve data from the Kepler Space Telescope, used for detecting exoplanets through transit events. The final CNN architecture, with hyperparameters set as (300, 200, 200, 100, 100), was identified as the best-performing model after evaluating multiple configurations. These results highlight the model's strengths and areas for improvement; while it excels at identifying false positives (low miss rate of 5%), its higher miss rate for the 'CONFIRMED' class (40%) suggests a need for better detection of true exoplanets. The AUC score of 0.91 further underscores the model's strong overall performance.