Partial Least Squares (PLS) Regression in Food Analysis

Why PLS?

Unlike ordinary least squares regression, PLS handles:

• Multicollinearity — spectral variables are highly correlated
• More variables than samples — typical in spectroscopy
• Noise reduction — by focusing on variance relevant to the response

Building a PLS Model

1. Collect reference samples with known property values
2. Acquire spectra for all samples
3. Apply spectral preprocessing (SNV, derivatives, etc.)
4. Split data into calibration and validation sets
5. Build the PLS model with cross-validation
6. Evaluate with RMSECV, R², and prediction errors