Biometric Analysis and Simulation Insight System (BASIS)

Modelling the Effects of Biometric Design Choices

Dr. Ted Dunstone — BixeLab / Biometix

Purpose

  • Demonstrate how biometric system design choices affect real-world outcomes.
  • Bridge technical metrics (accuracy, thresholds, modalities) with population-level impacts (enrolment success, fraud, inclusion).
  • Provide an interactive teaching and decision support tool for governments, standards bodies, and system integrators.

Why It Matters

  • Biometric performance depends on more than algorithms — it’s shaped by:
    • Population composition (rural vs. urban, ageing, diversity)
    • System configuration (thresholds, fusion rules)
    • Transaction scale and fraud pressure
  • Quantifying these factors clarifies trade-offs between:
    • Security (false accepts, fraud)
    • Usability (false rejects)
    • Inclusion (failure to enrol)

What the Simulator Does Now

Simulation Inputs

  • Population size, annual growth, transaction volume
  • Segment profiles (urban/rural, fraud rate, FTE rate)
  • Biometric modalities: Face, Fingerprint, Iris
  • Score distributions modelled with Gaussian or Poisson processes
  • Adjustable thresholds and fusion rules

Outputs and Visualisations

  • Score Histograms: genuine vs impostor distributions
  • ROC Curves: TPR vs FPR across thresholds
  • Threshold Tables: matching rates, FRR/FAR per setting
  • Annual Metrics:
    • People unable to enrol
    • Fraud attempts and successes
    • Population growth and transaction volume

Example Insights

  • Effect of raising the threshold on:
    • Security ↑, Inclusion ↓
  • Rural users experience slightly higher error rates → potential for bias visualisation
  • Fraud rate sensitivity: high FPR thresholds can amplify real-world exposure

Implementation Overview

  • Built with Dash + Mantine Components for interactive web UI
  • Modular simulation core in Python (NumPy / pandas)
  • Generates reproducible results (seeded random number generator)
  • Outputs interactive graphs and CSV tables for export

Potential Enhancements

Analytical Extensions

  • Add 1:N Identification mode (FPIR/TPIR modelling)
  • Multi-year cohort enrolment tracking
  • Correlated multimodal fusion (Gaussian copula)
  • Introduce cost functions and optimal operating point detection

User Experience

  • Interactive Dash dashboard with live parameter sliders
  • Scenario comparison mode for “what-if” analysis
  • Integration with real score data for calibration
  • Add equity and fairness dashboards by segment and demographic

Research and Policy Applications

  • Support ISO/IEC 19795 and 29794 test planning
  • Model impacts for digital identity inclusion programs (e.g., MOSIP, ID4D)
  • Train policymakers on trade-offs between accuracy, risk, and accessibility

Summary

Category Current Future
Modalities Face, Fingerprint, Iris Voice, Multi-modal Fusion
Stats Model Gaussian, Poisson Mixture Models, Empirical
Metrics ROC, FTE, Fraud Cost/Benefit, Fairness, Equity
Interface Dash Mantine Multi-scenario, Web-shareable
Purpose Demonstration & Education Policy Simulation & Standards Testing

Next Steps

  • Update documentation and planning
  • Enhance alpha build
  • Validate with empirical datasets
  • Package as open educational tool for biometric standards training
  • Connect to real operational data for parameter import
  • Publish under open-source licence (e.g., MIT)

Thank You

Biometric System Impact Simulator
📍 BixeLab | Biometix
💡 Exploring performance, inclusion & trust in identity systems