Automating Exoplanet Discovery: Enhancing Transit Method with Deep Learning and Transformers

The transit method, which detects exoplanets by observing periodic dips in stellar brightness, has significantly advanced our understanding of distant solar systems. This work enhances this method by integrating machine learning with TESS data to automate the identification of exoplanets and other celestial phenomena. Traditional models including k-means, random forest, and decision trees serve as baselines for performance comparison against two deep learning architectures: A Convolutional Neural Network (CNN) and a Transformer model. CNNs have excelled in recognizing patterns in data, such as transit signals in light curves, achieving higher accuracy and lower false positive rates than previous methods. Integrating Transformers marks a novel development in astronomical research. Pivotal in natural language processing and subsequently across domains, Transformers bring sophisticated attention mechanisms that enhance model interpretability particularly effective in unveiling weak signals that conventional models often overlook. This promises to improve current discovery techniques and deepen our understanding of the dynamics and compositions of distant planetary systems, paving the way for significant astronomical discoveries.