Cooperated Supervised and Semi-supervised Machine Learning for Identification of Exoplanet Habitability

The study of planetary habitability has gained widespread attention, and most existing studies focused on analyzing the habitability of a single planet based on a single feature, which makes it difficult to process a large amount of planetary data quickly. In this paper, we propose a machine learning-based identification method for efficiently distinguishing the habitability of a batch of planets. Firstly, a planet dataset comprising 5476 unlabeled records from the NASA Exoplanet Archive and 63 labeled entries with habitability from the Habitable Worlds Catalog is collected. Following that, a binary particle swarm optimization approach is used to select the most relevant features according to the 63 labeled planets. To address the missing values in the NASA data, next a standardized median imputation technique is applied. Two distinct methods, namely K-means clustering and distance-based filtering, are developed to label a subset of uninhabitable exoplanets by integrating the unlabeled 5476 and labeled 63 data points. Finally, KNN classifier and a semi-supervised label spreading classifier are trained and cooperated, contributing to the accomplishment of the final classification task. The experimental outcomes demonstrate the viability and effectiveness of the proposed method.