Multi-class Classification of Satellite Orbits for Database Quality Control
Document Type
Conference Proceeding
Publication Date
4-23-2025
Abstract
The Joint Spectrum Center (JSC) Equipment, Tactical, and Space (JETS) database contains 9,539 satellite records. When new data is ingested the satellite orbit type needs to be identified, which is currently a manual process. To save time, this work explores automating the process using machine learning. Several statistical machine learning and neural network models were developed and compared using the weighted averages of precision, recall, and F1 score metrics. The number of records used in training and testing was 1,024 with a 60/20/20 train, validation, and test split. Six orbital parameters were initially used to fit the models, but three parameters (the mean motion, eccentricity, and inclination) were most important in determining orbit type. A decision tree model with the three most important orbital parameters as inputs best identified the seven target orbit types. The weighted averages of the precision, recall, and F1 score on the test data were 0.991, 0.990, and 0.990 respectively. This compared favorably to the F1 metrics for a random classifier (0.106) and a model that always predicted the majority class (0.103).
Source Publication
Scientific Computing and Bioinformatics and Computational Biology (CSCE 2024). Communications in Computer and Information Science (CCIS), Volume 2258
Recommended Citation
Peterson, R., Wagner, T., Auclair, P., Langhals, B. (2025). Multi-class Classification of Satellite Orbits for Database Quality Control. In: Hodson, D.D., Grimaila, M.R., Arabnia, H.R., Deligiannidis, L., Wagner, T.J. (eds) Scientific Computing and Bioinformatics and Computational Biology. CSCE 2024. Communications in Computer and Information Science, vol 2258. Springer, Cham. https://doi.org/10.1007/978-3-031-85902-1_13
Comments
This conference paper is published by Springer and is available by purchase or subscription through the DOI link below.