[A late-type star spectra outlier data mining system].

Area of Science:

• Astronomy and Astrophysics
• Data Mining
• Stellar Classification

Context:

• M-type stars are crucial for understanding galactic structure and evolution due to their potential magnetic activity or rare properties.
• Identifying these unique M-type stars requires advanced data analysis techniques to overcome local biases in spectral characteristic lines.

Purpose:

• To develop and validate a data mining system for identifying outlier M-type stars from spectral data.
• To analyze the distribution of spectral characteristic lines and reduce dimensionality for outlier detection.

Summary:

• The system measures spectral line index distributions using sparse factor and sparsity coefficient, then discretizes and reduces data dimensionality.
• Particle Swarm Optimization (PSO) is employed to extract and identify local outlier subspaces.
• Experiments on SDSS M-type star spectral line index sets validate the system's feasibility and compare identified outliers with existing classifications.

Impact:

• Enables more accurate identification of rare M-type stars, crucial for follow-up observations.
• Enhances scientific research on galactic structure and evolution by providing a robust outlier detection method.
• Validates the effectiveness of the developed data mining system and its components, like PSO, in spectral analysis.