A Semi-Supervised Approach To Determine the Number of Exoplanets in a Given Planetary System Based on NASA Exoplanet Archive Data

Accurately determining the number of exoplanets in planetary systems is essential for gaining insight into planetary formation, the search for potentially habitable exoplanets, and the design of future exoplanet missions. Using data on exoplanets from the publicly accessible NASA Exoplanet Archive, unsupervised and supervised machine learning methods were applied to determine the number of exoplanets in a given planetary system. Clustering methods were applied in order to understand the multidimensional dataset and evaluate the potential efficacy of machine learning methods. The accuracies of models applied to this multiclass classification problem with imbalanced classes were obtained and reported, with several models achieving accuracies of over 90%. Applying algorithms including a decision tree, logistic regression, random forest, and support vector machine resulted in model accuracies comparable to similar studies where machine learning was applied to exoplanet data (Schanche et al. 2019). While the resulting model accuracies are on par with previous studies of exoplanet data, higher accuracies may be achieved with the incorporation of additional data from the James Webb Space Telescope (JWST) into this project (Schanche et al. 2019). Although these classification models were successful, due to the ever-changing nature of data on the number of planets present in planetary systems, cluster labels returned from clustering on the data were used as a target variable as a second approach. A significant increase in accuracy resulted, indicating that clustering methods, and specifically semi-supervised learning, hold great potential in identifying undetected exoplanets within known planetary systems.