Quantification of the Performance of Algorithms for spectra Baseline Correction

Sammanfattning: Spectroscopy serves as a vital tool in both scientific research and industrial applications. In spectral analysis, baseline correction is important in order to be able to efficiently extract essential features. Several algorithms for baseline correction have been developed, including Asls (Asymmetric Least Squares algorithm), arPLS (Asymmetrically Reweighted Penalized Least Squares algorithm), airPLS (Adaptive Iteratively Reweighted Penalized Least Squares algorithm), and MSBC (Multiple Spectra Baseline Correction algorithm). In this paper, a computational framework is devised to assess the efficacy of these four algorithms, based on principal component analysis, K-means clustering, confusion matrix and Silhouette analyses. Comprehensive computational experiments are conducted on synthetic and real Fourier transform infrared spectroscopy data. Drawing from our findings, we deduce that baseline correction significantly helps our spectral analysis by extracting information. Asls can be used to sort spectra into different clusters, and MSBC is able to attain consistent baselines and corrected spectra across all data. Moreover, we suggest potential avenues for refining baseline correction algorithms.