Semi-Supervised Classification of Stars, Galaxies, and Quasars Using K-means and Random-forest Approaches

Strategic Insights -

Research into astronomical classification has been a challenge due to the vast amount of data from modern surveys and the scarcity of labeled spectroscopic data. However, a new scalable and label-efficient method for astronomical classification has been proposed by researchers at the Institute of Advanced Studies Basic Science. This novel semi-supervised learning framework combines K-means clustering with random forest classification, showing promise in overcoming the limitations of fully supervised approaches. With the goal of developing a method that can efficiently classify astronomical objects, the researchers tested their approach on the CPz catalog, a dataset containing multi-survey photometric and spectroscopic data.

Key Takeaways:

  • The proposed semi-supervised learning framework leverages K-means clustering to partition unlabeled data into 50 clusters, and then propagates labels from spectroscopically confirmed centroids to 95% of cluster members.
  • The method trains a random forest on the expanded pseudo-labeled dataset and demonstrated robustness in high-dimensional feature spaces.
  • The proposed approach outperformed traditional color-cut techniques in classification accuracy, achieving F1 scores of 98.8%, 98.9%, and 92.0% for stars, galaxies, and quasars, respectively.
  • The method demonstrated superior label efficiency compared to prior work, making it a promising solution for astronomical classification when labeled data is limited.
  • The research showed that performance may be degraded in lower dimensional settings, indicating the need for further investigation.

Statistics:

  • The semi-supervised learning framework achieved F1 scores of 98.8%, 98.9%, and 92.0% for stars, galaxies, and quasars, respectively.
  • The method outperformed traditional color-cut techniques in classification accuracy.
  • The research demonstrated a 95% label propagation rate from spectroscopically confirmed centroids to cluster members.
  • The method showed robustness in high-dimensional feature spaces.

Sources:

  • Semi-supervised Classification of Stars, Galaxies and Quasars Using K-means and Random-forest Approaches. Astronomy & Astrophysics, 2025; 700.
  • Edp Sciences S A. 17, Ave Du Hoggar, Pa Courtaboeuf, Bp 112, F-91944 Les Ulis Cedex A, France.
  • V. Asadi, Institute of Advanced Studies Basic Science Iasbs, Dept. of Physics, Pob 113659161, Zanjan, Iran.
  • H. Haghi, Institute of Advanced Studies Basic Science.
  • A. H. Zonoozi, Institute of Advanced Studies Basic Science.
  • NewsRx. New Technology Study Findings Have Been Reported by Investigators at Institute of Advanced Studies Basic Science (Semi-supervised Classification of Stars, Galaxies and Quasars Using K-means and Random-forest Approaches). Journal of Engineering. October 20, 2025; p 2288.