[Automatic classification method of star spectra data based on manifold-based discriminant anaysis and Support Vector...

Area of Science:

• Astronomy
• Machine Learning
• Data Science

Context:

• Support Vector Machine (SVM) is a common tool in astronomical data analysis.
• Traditional SVM methods overlook class data distribution, limiting classification efficiency.
• Star spectra classification is crucial for understanding celestial objects.

Purpose:

• To propose a novel automatic classification method for star spectra data.
• To enhance classification accuracy by integrating manifold-based discriminant analysis (MDA) with SVM.
• To address the limitations of traditional SVM by considering data distribution and manifold structure.

Summary:

• A new method combines manifold-based within-class scatter (MWCS) and manifold-based between-class scatter (MBCS) from MDA with SVM.
• This approach optimizes the separating hyperplane to minimize MWCS and maximize MBCS.
• The method transforms the optimization problem to a quadratic programming (QP) dual form to find support vectors and decision functions.
• It effectively preserves the manifold structure of each class while considering inter-class information and distribution characteristics.

Impact:

• The proposed method demonstrates improved effectiveness in classifying star spectra compared to traditional approaches.
• It offers a more robust classification by accounting for both inter-class separability and intra-class compactness.
• Validation on Sloan Digital Sky Survey (SDSS) star spectra datasets confirms the method's efficacy.