Raw Raman spectral datasets of TEP, DIMP, and DMMP with a linear simulated dataset
Description
This dataset contains raw Raman spectral data of three organophosphorus compounds, namely triethyl phosphate (TEP), diisopropyl methylphosphonate (DIMP), and dimethyl methylphosphonate (DMMP), together with a linear simulated dataset used for methodological validation. The data support the study entitled "Prediction-space LOD calculation for nonlinear machine-learning Raman calibration: validation using RBF-SVR". The experimental Raman spectra were collected at different concentration levels to evaluate whether prediction-space definitions of quasi-blank prediction noise and low-concentration local sensitivity can be used for LOD calculation in nonlinear Raman calibration. The linear simulated dataset was generated to test the consistency of the proposed LOD calculation strategy under near-linear conditions using PLSR as a reference model. Each file contains a Number column, a Concentration column, and spectral-variable columns corresponding to Raman shifts in cm^-1. The experimental datasets use percentage concentration labels, whereas the simulated dataset uses fractional concentration labels in scientific notation. These data can be used to reproduce the Raman quantitative modeling, modified 3sigma/S LOD calculation, and Bootstrap-Monte Carlo LOD analysis described in the manuscript. Further details on file structure, concentration format, spectral range, and preprocessing are provided in README.txt.
Files
Steps to reproduce
The experimental Raman datasets were obtained from TEP, DIMP, and DMMP samples prepared at different concentration levels. Raman spectra were collected using a 1064 nm Raman spectrometer under the measurement conditions described in the associated manuscript. Each row in the experimental data files corresponds to one Raman spectrum, with the Number column indicating the spectrum index, the Concentration column indicating the sample concentration, and the remaining columns corresponding to Raman-shift variables in cm^-1. The linear simulated dataset was generated to mimic multivariate spectral data under near-linear concentration-response conditions. The concentration labels are expressed as fractional concentrations in scientific notation, and the spectral-variable columns correspond to Raman shifts in cm^-1. To reproduce the analysis described in the manuscript, users may apply Savitzky-Golay smoothing to the spectra, split the data into calibration and prediction sets, standardize the spectral variables and concentration labels using the calibration set, construct the RBF-SVR model, and then calculate the LOD using the modified 3sigma/S method and the Bootstrap-Monte Carlo procedure. Further details on file structure and concentration formats are provided in README.txt.
Institutions
- Aerospace Information Research InstituteBeijing, Beijing