[RVM supervised feature extraction and Seyfert spectra classification].

Area of Science:

• Astronomy and Astrophysics
• Machine Learning in Science

Context:

• Large-scale astronomical surveys (e.g., SDSS, 2dF, LAMOST) generate vast amounts of celestial spectral data.
• Automated spectral classification is crucial for analyzing this data deluge.
• Existing unsupervised feature extraction methods (PCA, WT, ANN, Rough Set) do not optimize for classification performance.

Purpose:

• To investigate the necessity and application of supervised feature extraction for celestial spectra classification.
• To introduce and evaluate Relevance Vector Machines (RVM) for supervised feature extraction and dimensional reduction.

Summary:

• This study highlights the limitations of unsupervised feature extraction methods in spectral classification.
• A supervised approach using Relevance Vector Machines (RVM), a Bayesian method, is proposed and applied to Seyfert spectra.
• RVM effectively extracts features and reduces data dimensionality based on classification capability, integrating training data and prior knowledge.

Impact:

• Demonstrates the superior performance of RVM in dimensional reduction and feature extraction for Seyfert classification.
• Provides a more effective method for automated celestial spectral classification, enhancing scientific discovery.
• Advances the field of machine learning applications in astronomical data analysis.