[Spectra Classification Based on Local Mean-Based K-Nearest Centroid Neighbor Method].

Area of Science:

• Astronomy and Astrophysics
• Machine Learning
• Data Science

Context:

• Accurate classification of celestial objects like stars, galaxies, and quasars is crucial for astronomical research.
• Existing methods such as K-nearest neighbors (KNN) and K-nearest centroid neighbors (KNCN) have limitations in handling complex spectral data.
• The Sloan Digital Sky Survey (SDSS-DR8) provides a vast dataset for testing and validating new classification algorithms.

Purpose:

• To introduce and evaluate a novel Local Mean-based K-nearest Centroid Neighbor (LMKNCN) technique for classifying stars, galaxies, and quasars (QSOS).
• To compare the performance of LMKNCN against traditional KNN and KNCN algorithms using real spectral data.
• To assess the effectiveness of LMKNCN, particularly in the identification of quasars.

Summary:

• The study proposes the LMKNCN technique, which leverages nearest centroid neighborhood principles and local mean vectors for classification.
• Experiments were conducted using spectral data from the SDSS-DR8, comparing LMKNCN with KNN and KNCN.
• LMKNCN demonstrated a higher or comparable rate of correct classification compared to the other methods, with notable improvements in quasar identification.

Impact:

• The LMKNCN algorithm offers a more accurate and robust approach to classifying celestial objects from spectral data.
• This improved classification accuracy has significant implications for astrophysical studies, including galaxy evolution and the understanding of quasars.
• The findings contribute to the advancement of machine learning applications in astronomical data analysis.