Dataset_Machine_Learning_Exoplanets_2024

The dataset used in this study consists of light curves collected by the Kepler telescope, totaling 5302 light curves, each with approximately 60,000 data points. The data were sourced from NASA's Exoplanet Archive, focusing on Kepler Objects of Interest (KOIs). Relevant columns such as kepid, koi_disposition, koi_period, koi_time0bk, koi_duration, and koi_quarters were selected. The Lightkurve library was utilized to extract the light curves, resulting in SAP (Simple Aperture Photometry) and PDCSAP (Pre-search Data Conditioning Simple Aperture Photometry) fluxes. Due to its precision, PDCSAP flux was used for exoplanet detection. Normalization was performed to standardize the light curves, missing data were addressed through linear interpolation, and outliers were removed using a 2-standard deviation threshold. An extensive experimental evaluation involving 16 algorithms with different parameter settings determined that the LightGBM algorithm demonstrated the best performance, achieving an accuracy of 82.92%. The results highlight the effectiveness of LightGBM for exoplanet classification. For more details, refer to the article: Macedo, B. H. D., & Zalewski, W. (2024). Automated Light Curve Processing for Exoplanet Detection Using Machine Learning Algorithms. Rev. Bras. de Iniciação Científica (RBIC), IFSP Itapetininga, 11, e024021, 1-27. Access the code on Website: https://brunohdmacedo.engineer/project.html. "As a key result of the experimental evaluation, the LightGBM algorithm achieved the best performance with an accuracy rate of 82.92%".