2026AJ....171...28H

Exoplanet detection is an active research area in astronomy. As the volume of astronomical data increases, there is a need for methods that can automate the data processing pipeline efficiently. This study examines how classical machine learning algorithms, can classify exoplanet candidates without using deep learning models. Our comparative analysis of different machine learning techniques shows that classical algorithms can perform at a level comparable to advanced deep learning methods. By utilizing the time series analysis library tsfresh, we were able to extract 740 features from each light curve, effectively capturing its key statistical, temporal, and spectral characteristics. Among the evaluated machine learning techniques, XGBoost and XGBoost calibrated demonstrated the highest predictive performance. XGBoost achieved an accuracy of 89%, followed by histogram gradient boosting at 87%. Sensitivity and precision metrics remained consistent across models, with F1-scores ranging from 82% to 89%. We further employed receiver operating characteristic and precision-recall curves to evaluate model performance, demonstrating their ability to distinguish between positive instances while minimising false positives. Finally, our calibrated XGBoost-based model revalidates 14 exoplanet candidates from NASA's Exoplanet Archive, which were validated against Kepler Robovetter confidence scores and benchmarked with known exoplanet discoveries.