Stochastic generalization models learn to comprehensively detect volatile organic compounds associated with foodborne...

Area of Science:

• Analytical Chemistry
• Food Science
• Biotechnology

Background:

• Food safety relies on detecting pathogens and contaminants.
• Traditional methods like culturing are slow and labor-intensive.
• Rapid, accurate detection systems are needed for real-time monitoring.

Purpose of the Study:

• To develop a system combining Raman spectroscopy and machine learning for volatile organic compound (VOC) detection.
• To precisely identify and quantify VOCs associated with foodborne pathogens in complex mixtures.
• To offer a rapid, on-site alternative to conventional food safety testing methods.

Main Methods:

• A remote fiber-optic Raman probe was used to collect spectral data from 42 distinct VOC mixtures.
• Machine learning models (MLP, random forest, XGBDT) were trained on 1445 Raman spectra.
• Classification and regression analyses were performed on diluted and undiluted VOC samples.

Main Results:

• Machine learning models achieved over 90% accuracy in classifying pure VOCs.
• The system reliably identified mixtures of up to six VOCs at 0.25% concentration (400-fold dilution).
• Regression models accurately predicted VOC concentrations down to 1% (100-fold dilution) with R² > 0.82.

Conclusions:

• The combined Raman spectroscopy and ML system offers a powerful tool for food safety.
• This approach enables rapid, real-time detection and analysis of foodborne pathogen indicators.
• It provides a valuable solution for on-site food safety assessments and quality control.