Machine learning enabled surface-enhanced Raman spectroscopy quantitative analysis for food safety monitoring

Area of Science:

• Analytical Chemistry
• Food Safety Science
• Spectroscopy

Background:

• Crystal violet (CV) is a prohibited dye misused in aquaculture, posing toxicological risks and threatening food safety.
• Accurate and sensitive detection methods for CV residues in fish are crucial for public health.

Purpose of the Study:

• To develop a fast and highly sensitive strategy for determining CV residues in fish matrices.
• To compare the performance of different machine learning algorithms for CV quantification.

Main Methods:

• Utilized silver nanocubes as a surface-enhanced Raman spectroscopy (SERS)-active platform for signal amplification.
• Employed three regression algorithms: Partial Least Squares (PLS), Support Vector Machine (SVM), and Random Forest (RF) for quantification.
• Evaluated algorithm performance based on prediction accuracy (R²), mean absolute error, and stability.

Main Results:

• The Random Forest (RF) model demonstrated superior predictive capability with a test-set R² of 0.9926 and a low mean absolute error of 0.0046 ppm.
• Partial Least Squares (PLS) showed limitations due to linear assumptions and overfitting (R² = 0.9529).
• Support Vector Machine (SVM) performed well at lower concentrations but was sensitive to noise.

Conclusions:

• The developed SERS-coupled RF approach offers a reliable and sensitive method for detecting crystal violet residues in fish.
• This strategy enhances food safety by enabling accurate monitoring of prohibited dye contamination.
• The RF model's ability to integrate broad spectral information ensures stable predictions even at high analyte concentrations.