Optimized variable selection and machine learning models for olive oil quality assessment using portable near...

Area of Science:

• Agricultural Science
• Analytical Chemistry
• Food Science

Background:

• Olive oil quality is crucial for the Mediterranean diet, but traditional chemical tests are costly and time-consuming.
• Accurate assessment of olive oil quality parameters like acidity, K232, and K270 is essential due to market demand and price variations.

Purpose of the Study:

• To investigate the efficacy of near-infrared spectroscopy (NIRS) for predicting key olive oil quality parameters.
• To develop a faster and more economical method for olive oil quality assessment compared to traditional chemical analyses.

Main Methods:

• Collected 200 Moroccan olive oil samples across three quality categories (extra virgin, virgin, ordinary virgin).
• Utilized a portable NIR spectrometer to obtain spectral data and conducted chemical analyses following international standards.
• Applied Partial Least Squares Regression (PLSR) with variable selection algorithms (e.g., iPLS) to correlate spectral data with chemical parameters.

Main Results:

• NIRS combined with iPLS achieved excellent prediction for acidity (R²=0.94, RPD=4.2).
• Moderate prediction performance was observed for K232 and K270 parameters (R² between 0.60-0.75).
• Principal Component Analysis (PCA) enabled differentiation between olive oil quality groups, with variable selection enhancing predictive accuracy.

Conclusions:

• NIRS offers a viable, efficient, and accurate method for predicting olive oil acidity and other quality indicators.
• Variable selection techniques significantly improve the predictive power of NIRS models for olive oil analysis.
• This approach provides a cost-effective alternative for rapid quality assessment of olive oil using portable spectrometers.