Skip to content

Tutorial: Oil Authentication

Purpose: Authenticate edible oils (Raman/FTIR) using ratiometric features and validation.

Audience: Food scientists, QA engineers, chemometricians.

Time: 30–45 minutes.

Prerequisites: FoodSpec installed, CSV or HDF5 data with oil types and spectra.

Overview

This tutorial demonstrates how to authenticate edible oils using vibrational spectroscopy. You will:

  • Load spectral data (CSV or HDF5) with oil labels
  • Apply preprocessing (baseline correction, normalization)
  • Extract ratiometric features (e.g., C=O/C=C band ratios)
  • Train and validate a classifier with nested cross-validation
  • Generate publication-ready figures and metrics

Example data: examples/data/oils.csv (Raman spectra, oil types, batch info)
Protocol: examples/protocols/oil_basic.yaml (baseline ALS, SNV normalization, Random Forest)

Why it matters (theory)

Authenticating oils protects against adulteration and ensures quality. Discriminative ratios (e.g., carbonyl/unsaturation bands) separate oil types. Validation (batch-aware CV) checks generalization; see rq_engine_theory.md and chemometrics_and_ml_basics.md.

CLI workflow

foodspec-run-protocol \
  --input examples/data/oils.csv \
  --protocol examples/protocols/oil_basic.yaml \
  --output-dir runs/oil_basic_tutorial

foodspec-publish runs/oil_basic_tutorial/<timestamp> --fig-limit 6
Check figures/ for confusion matrix and discriminative barplots; tables/ for discriminative and minimal panel summaries.

Example figures (from run bundle)

Confusion matrix Top discriminative ratios Minimal panel accuracy Stability

How to read the results

  • Confusion matrix/balanced accuracy: off-diagonals show misclassifications; balanced accuracy corrects imbalance.
  • Discriminative ratios/effect sizes: look for high importance or low FDR-corrected p-values.
  • Minimal marker panel: smallest feature set meeting target accuracy—useful for QA panels.
  • Stability (CV/MAD): lower CV/MAD indicates more reproducible markers.