Skip to content

Oil Authentication: Supervised Classification

Level: Beginner → Intermediate
Runtime: ~10 seconds
Key Concepts: Classification, cross-validation, confusion matrices, model discrimination

What You Will Learn

In this example, you'll learn how to: - Load spectroscopy data from CSV files - Train a classifier to distinguish oils by type - Evaluate model performance with cross-validation - Interpret confusion matrices and classification metrics - Visualize data structure using dimensionality reduction (PCA)

After completing this example, you'll understand the workflow for any classification problem in FoodSpec (fraud detection, quality assessment, authenticity verification).

Prerequisites

  • Basic Python knowledge
  • Familiarity with NumPy arrays and Pandas DataFrames
  • Understanding of supervised learning concepts (train/test, classification)
  • numpy, pandas, matplotlib, scikit-learn installed

Optional background: Read Chemometrics & ML Basics

The Problem

Real-world scenario: You're a food manufacturer testing whether your olive oil supplies are authentic. You have reference spectra for virgin olive oil, processed olive oil, and two adulterants (sunflower, canola). Can you build a classifier to automatically detect fake oils?

Data: Raman spectra (intensity vs. wavenumber) for 8 samples across 4 classes.

Goal: Train a model, evaluate with cross-validation, interpret results.

Step 1: Load Data

import numpy as np
import pandas as pd
from pathlib import Path

# Load synthetic oil spectra (8 samples × 1500 wavelengths)
data = pd.read_csv("examples/data/oil_synthetic.csv", index_col=0)
X = data.drop("OilType", axis=1).values
y = data["OilType"].values

print(f"Data shape: {X.shape}")  # (8, 1500)
print(f"Oil types: {np.unique(y)}")  # ['CanOil' 'OliveOil' 'ProcessedOl' 'SunflowerOil']

What's happening: - The CSV contains 8 spectra and their oil type labels - X contains the spectral intensities (predictors) - y contains the oil type labels (targets)

Step 2: Train with Cross-Validation

from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import cross_validate

# Initialize classifier
clf = RandomForestClassifier(n_estimators=50, random_state=42)

# Evaluate with 5-fold cross-validation
scores = cross_validate(
    clf, X, y, 
    cv=5, 
    scoring=["accuracy", "precision_macro", "recall_macro", "f1_macro"]
)

print(f"CV Accuracy: {scores['test_accuracy'].mean():.3f} ± {scores['test_accuracy'].std():.3f}")
print(f"CV F1: {scores['test_f1_macro'].mean():.3f} ± {scores['test_f1_macro'].std():.3f}")

What's happening: - RandomForestClassifier is trained on 4/5 of the data, tested on 1/5 - This repeats 5 times, rotating which fold is held out - We compute accuracy, precision, recall, F1 across all folds - Metrics near 1.0 indicate good discrimination

Step 3: Visualize Performance & Structure

from sklearn.model_selection import cross_val_predict
from sklearn.metrics import confusion_matrix
import matplotlib.pyplot as plt

# Get predictions from cross-validation
y_pred = cross_val_predict(clf, X, y, cv=5)
cm = confusion_matrix(y, y_pred)

# Plot confusion matrix
fig, ax = plt.subplots(figsize=(6, 5))
im = ax.imshow(cm, cmap="Blues")
ax.set_xticks(range(len(np.unique(y))))
ax.set_yticks(range(len(np.unique(y))))
ax.set_xticklabels(np.unique(y), rotation=45)
ax.set_yticklabels(np.unique(y))
ax.set_xlabel("Predicted")
ax.set_ylabel("Actual")
ax.set_title("Oil Classification: Confusion Matrix")

# Add counts
for i in range(cm.shape[0]):
    for j in range(cm.shape[1]):
        text = ax.text(j, i, cm[i, j], ha="center", va="center", color="w")

plt.tight_layout()
plt.savefig("oil_auth_confusion.png", dpi=150, bbox_inches="tight")
plt.show()

Interpretation: - Diagonal elements = correct classifications - Off-diagonal = misclassifications - Perfect classifier = only non-zero diagonals - This plot reveals which oils are confused with others

Full Working Script

See the production script with enhanced documentation, output directory management, and additional analysis:

📄 examples/oil_authentication_quickstart.py – Full working code (35 lines)

Generated Figure

Confusion Matrix

Figure interpretation: - Rows = true oil types - Columns = predicted oil types - Perfect classification: all counts on diagonal - Model performance: assess which oils are well-distinguished

Key Takeaways

✅ Classification workflow: Load → Train → Cross-validate → Evaluate
✅ Cross-validation: Prevents overfitting by rotating train/test splits
✅ Confusion matrix: Shows misclassification patterns, not just accuracy
✅ Metrics matter: Precision/recall assess class-specific performance

Real-World Applications

  • 🌾 Olive oil authentication: Detect counterfeit high-value oils
  • 🍯 Honey fraud detection: Distinguish pure from adulterated honey
  • 🧈 Butter authenticity: Identify margarine substitution
  • 🥛 Milk origin verification: Grass vs. grain-fed dairy

Next Steps

  1. Try it: Modify the classifier (e.g., use SVM, Logistic Regression)
  2. Explore: Change cross-validation folds (cv=10) and observe variance
  3. Learn more: Read Classification & Regression
  4. Advance: See Oil Authentication Workflow for complete domain example

Interactive Notebook

For step-by-step exploration with visualizations:

📓 examples/tutorials/01_oil_authentication_teaching.ipynb

Figure provenance