講演情報
[15p-S2_201-2]Noise Dependence of PCA-Based Feature Extraction in Fe-57 Mössbauer Spectra
〇Sonia Sharmin1, Chiharu Mitsumata1 (1.Tsukuba Univ.)
キーワード：
Mossbauer spectroscopy、principal component analysis、noise effect
Principal component analysis (PCA) is widely used for feature extraction and dimensionality reduction in the analysis of any kind of spectra acquired under limited counting statistics. In this study, we examine the noise dependence of PCA-based feature extraction in Fe-57 Mössbauer spectra, with particular emphasis on magnetically split (sextet) line shapes. Ensembles of synthetic magnetic, paramagnetic, and mixed spectra are analyzed over a controlled range of noise levels, and changes in PCA loading vectors and variance distributions are evaluated. With increasing noise, the dominant principal components exhibit gradual modifications rather than an abrupt loss of low-dimensional structure, while spectral variance is progressively redistributed toward higher-order components. In magnetic spectra, noise mainly affects variance directions associated with hyperfine splitting and relative line intensities. These results indicate that noise-induced changes in PCA structure are directly related to physically meaningful spectral features. The proposed analysis provides a model-independent framework for assessing the robustness of PCA-based representations when conventional fitting methods become unreliable.