Application of Raman spectroscopy and Machine Learning algorithms for fruit distillates discrimination

Area of Science:

• Analytical Chemistry
• Chemometrics
• Spectroscopy

Background:

• Distillate authentication is crucial for quality control and consumer protection.
• Traditional methods for differentiating distillates can be time-consuming and lack precision.
• Novel analytical techniques are needed for rapid and accurate distillate analysis.

Purpose of the Study:

• To explore the efficacy of Raman spectroscopy coupled with Machine Learning (ML) for differentiating distillates.
• To determine the potential of specific Raman spectral regions for classification.
• To assess the accuracy of the combined approach for trademark, geographical, and botanical origin discrimination.

Main Methods:

• Utilized Raman spectroscopy to acquire spectral data from various distillates.
• Applied Machine Learning algorithms to analyze spectral data and build classification models.
• Focused on two key spectral regions (200-600 cm⁻¹ and 1200-1400 cm⁻¹) for enhanced discrimination.
• Evaluated model performance based on accuracy and misclassification rates.

Main Results:

• The combined Raman spectroscopy and ML approach achieved high accuracy in distillate differentiation.
• Trademark fingerprint differentiation reached 95.5% accuracy.
• Geographical discrimination within the Transylvania region yielded 90.9% accuracy.
• Botanical origin differentiation was most effective within specific producing entities due to trademark dominance.

Conclusions:

• Raman spectroscopy and ML offer a powerful, accurate, and rapid method for distillate authentication.
• The technique shows significant potential for verifying trademark, geographical, and botanical origin.
• This pilot study demonstrates a promising non-destructive analytical tool for the spirits industry.