Stellar Spectral Subclass Classification Based on Locally Linear Embedding

Abstract

Abstract Locally linear embedding (LLE) is a recently developed dimension reduction technique. In this paper, we describe how we applied LLE to the stellar subclass classification. We found that LLE classifies the objects with different physical characteristics correctly. We then compared the performance of LLE with that of principal component analysis (PCA) in spectral classification, and found that LLE does better than PCA. We tested the robustness of LLE against the changing of signal-to-noise ratios (SNRs), and found that the performance of LLE is affected by two factors: changing of SNRs and the range of SNRs of the spectra data set. We also studied the variation of LLE parameters, and found that the experiment results are affected by the parameter variation, but not sensitive. Finally, using LLE, we located those objects misclassified by the Sloan Digital Sky Survey pipeline, and estimated its accuracy in classifying stellar subclasses.