Rapid Sweetener Detection Achieved Through Raman Spectroscopy and Machine Learning

A recent study examined a new method that can rapidly and accurately detect sweeteners in various food products. This study, which was published in the journal Infrared Physics & Technology, demonstrated how Raman spectroscopy, when combined with a machine learning (ML) algorithm, can accurately identify sweeteners used in food production (1). This study was led by Yingaridi Bu, a researcher at Heilongjiang University, and the findings indicate how spectroscopy has a future role to play in real-time ingredient verification.

What are the concerns right now in food safety?

Currently, the food industry is receiving more scrutiny. At the federal level, the ingredients in food are being closely examined, with the Department of Health and Human Services recommending bans on specific artificial additives and ingredients (2). To help preserve food, as well as make it taste better, artificial additives are being used more frequently in food products. The issue, though, is that many of these additives are harmful to human health. Artificial sweeteners, for example, can pose risks when consumed excessively, causing potential damage to the gastrointestinal tract (1).

Therefore, this study examined a better way to accurately and timely identify these sweeteners in consumer products to better inform consumers and safeguard human health.

What limitations does Raman spectroscopy have, and how can ML algorithms resolve these concerns?

Previous studies have already looked into how Raman spectroscopy performs when analyzing the chemical composition of food additives (3,4). However, the process of labeling or classifying detected sweeteners has typically required manual interpretation of spectral data, a time-consuming and error-prone step that limits the method’s use in rapid-response scenarios such as quality control during food processing or customs inspections (1).

ML algorithms can automate the spectral data interpretation step, resulting in reduced analysis times. In the study, the research team incorporated the Random Forest (RF) algorithm into the sweetener classification process. The algorithm was trained to identify three commonly used sweeteners: sucrose, cyclamate, and glucose (grape sugar) (1).

To prepare the Raman spectra for ML analysis, the researchers applied advanced signal preprocessing techniques, including adaptive iteratively reweighted penalized least squares (airPLS) for baseline correction and Savitzky-Golay filtering for noise reduction (1). These steps enhanced the clarity of spectral data, enabling the algorithm to more effectively distinguish between the molecular signatures of the sweeteners.

How did their method perform? What limitations remain?

The researchers found through their laboratory tests that their method was nearly perfect in identifying the three sweeteners. In addition, the detection times were also noteworthy, with per sample taking only 5–6 seconds (1).

However, the research team also acknowledged a few challenges encountered in the study, which future studies could explore in greater detail. For example, because Raman spectral data is sensitive to experimental variables such as laser intensity, acquisition time, and the physical characteristics of the sample surface, the research team opted to standardize these parameters across all tests (1). Additionally, the effectiveness of the RF algorithm is closely tied to the quality and diversity of the training dataset (1). To improve model generalizability, the researchers expanded the data set and used cross-validation techniques to bolster performance (1).

Another challenge faced in the study relates to distinguishing sweeteners. Some of the sweeteners have similar molecular structures, which can produce overlapping spectral features (1). The researchers suggest that future studies might examine using spectroscopic techniques such as Fourier-transform infrared (FT-IR) or near-infrared (NIR) spectroscopy to see if either resolves this issue (1). Furthermore, the researchers also suggested that future studies can apply dimensionality reduction techniques like principal component analysis (PCA) or more sophisticated deep learning models to the analysis (1). By doing so, they may be able to achieve improved accuracy and resilience against spectral noise.

References

1. Ding, Y.; He, X.; Zhang, R.; Wu, H.; Bu, Y. Random forest-assisted Raman Spectroscopy and Rapid Detection of Sweeteners. Infra. Phys. Technol. 2025, 148, 105871. DOI: 10.1016/j.infrared.2025.105871
2. U.S. Food and Drug Administration, HHS, FDA to Phase Out Petroleum-Based Synthetic Dyes in Nation’s Food Supply. FDA.gov. Available at: https://www.fda.gov/news-events/press-announcements/hhs-fda-phase-out-petroleum-based-synthetic-dyes-nations-food-supply (accessed 2025-07-09).
3. He, S.; Xie, W.; Zhang, W.; et al. Multivariate qualitative analysis of banned additives in food safety using surface-enhanced Raman scattering spectroscopy. Spectrochimica Acta Part A: Mol. Biomol. Spectrosc. 2015, 137, 1092–1099. DOI: 10.1016/j.saa.2014.08.134
4. Podstawka, E.; Swiatlowska, M.; Borowiec, E. Food additives characterization by infrared, Raman, and surface-enhanced Raman spectroscopies. J. Raman Spectrosc. 2007, 38 (3), 356–363. DOI: 10.1002/jrs.1653