Locally linear embedding: dimension reduction of massive protostellar spectra

TL;DR

This paper demonstrates that locally linear embedding (LLE) effectively reduces the dimensionality of massive protostellar spectra, outperforming PCA and Isomap in spectral classification tasks.

Contribution

It introduces the application of LLE to large spectral datasets of protostars and compares its performance with PCA, Isomap, and Hessian LLE.

Findings

LLE outperforms PCA and Isomap in spectral classification.

LLE is a valuable tool for analyzing large spectral datasets.

Limitations of LLE are acknowledged but it remains superior for classification.

Abstract

We present the results of the application of locally linear embedding (LLE) to reduce the dimensionality of dereddened and continuum subtracted near-infrared spectra using a combination of models and real spectra of massive protostars selected from the Red MSX Source survey database. A brief comparison is also made with two other dimension reduction techniques; Principal Component Analysis (PCA) and Isomap using the same set of spectra as well as a more advanced form of LLE, Hessian locally linear embedding. We find that whilst LLE certainly has its limitations, it significantly outperforms both PCA and Isomap in classification of spectra based on the presence/absence of emission lines and provides a valuable tool for classification and analysis of large spectral data sets.