Exoplanet detection using ensemble learning: A comparative study with gradient boosting as the optimal classifier

The detection of exoplanets is essential for advancing our knowledge of planetary systems beyond our solar system. The performance of several machine learning models to detect exoplanet candidates in the Cumulative Kepler Objects of Interest (KOI) is evaluated in this study. We applied ensemble learning techniques, including Random Forest, Gradient Boosting, Voting Classifier, Stacking Classifier and feature selection approaches, ANOVA F test, Mutual Information Gain, Recursive Feature Elimination (RFE). The best performance among the models was obtained by Gradient Boosting with a mean accuracy of 99.80%, and a mean F1 score of 99.81%, which surpassed the other classifiers. Gradient Boosting was not only the most accurate classifier, but it also achieved the highest F1 score, just a little low behind the Stacking and Voting Classifiers. This result shows the effectiveness of Gradient Boosting for exoplanet detection and demonstrates the merit of combining ensemble learning with high dimensional feature selection techniques to improve classification accuracy.