EXO-AI: A Multi-Mission LightGBM Model for Exoplanet Classification

EXO-AI is an Artificial Intelligence model that uses data from three NASA satellite missions, Kepler (KOI), K2, and TESS, harmonized into one unified dataframe to detect exoplanets – planets outside our solar system. While effective, traditional methods for detecting exoplanets, like the transit method and radial velocity techniques, are costly, time-consuming and require manual verification, which can take up to months. Recent advances in Artificial Intelligence have made exoplanet detection faster, but most existing models are trained on data from a single satellite mission, making cross-mission generalization more difficult. EXO-AI addresses this by leveraging a LightGBM model trained on 18 shared features from KOI and K2 data, with TESS data used for external testing despite reduced feature availability. The model achieved strong results on KOI + K2 testing data (F1 = 0.934, PR-AUC = 0.982) but weaker performance on TESS (F1 = 0.282, PR-AUC = 0.617). These results show that multi-mission integration improves detection accuracy within shared feature spaces but also emphasize the need for standardized features to achieve consistent, cross-mission reliability. Finally, an interactive Streamlit web app was developed where users can input astrophysical features and instantly get a prediction score, using the model, indicating the likelihood of a real exoplanet.