ATR-FTIR spectra, physicochemical properties, and machine-learning code for sustainable aviation fuel blends
Description
This dataset supports research on the prediction of physicochemical properties of sustainable aviation fuel (SAF) blends using attenuated total reflectance Fourier-transform infrared (ATR-FTIR) spectroscopy and machine learning. The dataset contains raw ATR-FTIR absorbance spectra of 50 fuel samples, with 2,385 spectral variables covering approximately 3997–599 cm⁻¹. Reference measurements include density, 10% and 90% distillation temperatures (T10 and T90), and kinematic viscosities at −20 and −40 °C for all 50 samples. Freezing-point measurements are available for 33 samples. Fuel-blend composition information and sample identifiers are provided to support data interpretation and sample matching. Where included, Python scripts and outer-fold out-of-fold prediction results document the machine-learning evaluation workflow. Missing freezing-point measurements are retained as missing values rather than imputed.
Files
Steps to reproduce
The ATR-FTIR spectra and reference property measurements are aligned using the sample identifiers provided in the data files. The Python modeling scripts implement spectral preprocessing, feature selection, regression modeling, and fivefold outer and fivefold inner nested cross-validation. Out-of-fold predictions are used to calculate R², RMSE, and MAE for each property. The README provides software dependencies, data structure, execution instructions, and details of the evaluation configuration.
Institutions
- Civil Aviation University of ChinaTianjin, Tianjin