ormat varies with the spectrum across experiments and samples.

efore leads to the difficulty when analysing the spectrometric data

ting chemicals or molecules from a spectrum. This difficulty has

resulted in the development of various algorithms for baseline

n.

are mainly three types of baseline estimation algorithms. The

is to fit a smoothing spline to a spectrum [Hastie and Tibshirani,

he fitting is based on a pre-defined model structure, i.e., the

g order or the smoothing degree. These algorithms have a typical

e of simplicity and therefore have been widely used in a variety

ations for discovering chemicals or molecules from spectra data

the Raman spectra [Shusterman, et al., 2000; Fonseca, et al.,

hulze, et al., 2019; Monteiro, et al., 2020]. However, these

ms have a typical limitation because a model structure must be pre-

nd the baseline removal performance is closely related with the

design of a smoothing degree [Froning, et al., 1988]. These

ms also have a problem to effectively deal with some complicated

structures [Perperoglou, et al., 2019].

econd type of baseline estimation algorithms is to use the wavelet

g technique, or the wavelet decomposition and integration

[Price, et al., 2008; Restrepo-Agudelo, et al., 2017]. The

assumes that a spectrum has a typical property that a high wave

corresponds to a chemical peak (or a signal) and a low wave

represents a baseline value. Therefore two types of spectrometric

can be separated by decomposing two types of wave intensities.

elet approach has also been applied to the Raman spectrometric

ang, et al., 2005; Alimova, et al., 2019]. However, this approach

problem that an optimised threshold must be carefully determined

the low and high wave intensities [Hu, et al., 2007].

hird type of algorithms for baseline estimation is called the least

error approach (LSE). LSE has been the mostly and widely

one for the estimation of a baseline in a spectrum. This is mainly

of its flexibility though it has also a problem of model selection

sation. Both the classical and the LSE variants have been used to