Data augmentation in food science: Synthesising spectroscopic data of vegetable oils for performance enhancement

Abstract

Generating more accurate, efficient, and robust classification models in chemometrics, able to address real‐world problems in food analysis, is intrinsically related with the amount of available calibration samples. In this paper, we propose a data augmentation solution to increase the performance of a classification model by generating realistic data augmented samples. The feasibility of this solution has been evaluated on 3 main different experiments where Fourier transform mid infrared (FT‐IR) spectroscopic data of vegetable oils were used for the identification of vegetable oil species in oil admixtures. Results demonstrate that data augmented samples improved the classification rate by around 19% in a single instrument validation and provided a significant 38% improvement in classification when testing in more than 10 different spectroscopic instruments to the calibration one.

Application of a Hybrid Fusion Classification Process for Identification of Microplastics Based on Fourier Transform Infrared Spectroscopy

“…Recent work demonstrated the utility augmenting simulated spectra to represent possible variations in real spectra. 72 Such an approach is ongoing in the laboratory. An alternative approach also undergoing studies in the laboratory is classification updating (also referred to as transfer learning).…”

“…A supplementary approach to improve microplastic identification by classification is to increase the number of reference spectra using additional simulated samples expressed as various linear combination of current spectra, 72 simulated weathered samples under additional simulated conditions, and environmental samples in order to form base reference classes spanning a larger variety spectral distortions. Classification updating is another strategy that has proven to be successful in other classification situations.…”

Near-infrared spectroscopy-based methods for quantitative determination of active pharmaceutical ingredient in transdermal gel formulations

“…[28] Simulated spectra based on the measured datasets were used to confirm the results of PLSR models based on original datasets. [17,18] The five independent G1-G5 datasets were also applied to build PLSR models and these independent models were tested with the original spectra. Mean plots of RMSE and RPD of PLSR models with confidence intervals (±0.95) are shown in Fig.…”

“…These kinds of data augmentation can facilitate the optimization of modelbuilding techniques when the low availability and high costs of samples may limit experimental sample sizes. [17,18] Notwithstanding, it should be taken into account that the models thus constructed only represent the variability of the samples being tested. For later use, it is recommended to increase the number of samples and rebuild calibrations.…”

Continuous statistical modelling in characterisation of complex hydrocolloid mixtures using near infrared spectroscopy

“…Moreover, a differentiation of the carrageenan+gum+excipient samples between carrageenan+gum+KCl and carrageenan+gum+MD samples is also noticeable. Regarding the testing phase, before the same data pre-treatment, linear interpolation was applied to the test spectra in order to get the desirable number of variables (n=1050) since they were acquired with a spectrometer with different spectral resolution [21,22] to reduce overfitting to the NIR calibration set.…”