Human Breathprints as Biometric Identifiers
Assessing Uniqueness and Reliability — investigated whether human "breathprints" can act as a reliable biometric identifier using machine learning on volatile organic compound (VOC) breath data.
Research question
Can VOC patterns in exhaled breath uniquely and reliably identify an individual despite natural variability caused by diet, health, environment, and time?
- Assess uniqueness via patterns across individuals
- Reduce high-dimensional VOC data into informative components
- Prevent overfitting given limited samples
- Evaluate classification performance using standard metrics
Dataset & constraints
Breath samples from 11 subjects collected across multiple days. VOCs measured using SESI-MS / Q-TOF mass spectrometry.
- High dimensionality: many VOC features
- Small sample size: risk of overfitting
- Variability: VOCs fluctuate over time
Pipeline (Iteration 1)
Goal: reduce dimensionality early, then select features.
- StandardScaler → PCA → LDA
- Lasso / Ridge / ElasticNet feature selection
- Classifiers: Logistic Regression, SVM, Random Forest
Pipeline (Iteration 2)
Goal: test whether ordering matters (feature selection first).
- StandardScaler → Lasso / Ridge / ElasticNet
- LDA after selection (maximise class separability)
- Same classifiers: Logistic Regression, SVM, Random Forest
KEY FINDING: ordering matters
Changing the sequence of dimensionality reduction and feature selection significantly affects classification outcomes. Iteration 2 (feature selection → LDA) produced a major improvement in Logistic Regression performance.
Iteration 1 — Results
| Model | Accuracy | F1 |
|---|---|---|
| Logistic Regression | 25.64% | 24.46% |
| SVM | 30.77% | 33.03% |
| Random Forest | 33.33% | 32.01% |
Best accuracy: Random Forest (~33%) — problem is hard under this pipeline.
Iteration 2 — Results
| Model | Accuracy | F1 |
|---|---|---|
| Logistic Regression | 56.41% | 55.99% |
| SVM | 41.03% | 44.74% |
| Random Forest | 38.46% | 32.01% |
Logistic Regression improves dramatically — feature space shape matters more than classifier choice.
What the results suggest
- Promise: Breathprints contain identifying information
- Hard problem: High-dimensional signals + temporal variability
- Representation matters: Better feature space unlocks stronger results
Limitations & future work
- Scale: Larger labelled datasets needed
- Stability: Evaluate across longer time windows
- Methods: Non-linear reduction, stronger ensembles
- Validation: Test on unseen cohorts / external datasets