Catalog
method#Analytics#Software Engineering#Architecture

Agent-Based Modeling (ABM)

ABM models autonomous agents and their interactions to study emergent phenomena in complex systems.

Agent-Based Modeling (ABM) is a simulation-based method for studying complex systems by modeling autonomous agents and their interactions.
Established
Medium

Classification

  • Medium
  • Technical
  • Design
  • Intermediate

Technical context

Data pipelines for input and calibration data (CSV, APIs)Visualization and dashboard tools for presenting resultsAnalytical tools for sensitivity and uncertainty analyses

Principles & goals

Model agents explicitly and minimally: represent only relevant attributes and rules.Validate against data and perform plausibility checks.Develop incrementally and document experiments systematically.
Discovery
Enterprise, Domain, Team

Use cases & scenarios

Compromises

  • Overinterpretation of model results without adequate validation.
  • Mis-calibration can lead to misleading recommendations.
  • Black-box models reduce explainability and stakeholder acceptance.
  • Start with simple models and extend incrementally.
  • Keep scripts and experiments reproducible and documented.
  • Engage stakeholders early and make model assumptions transparent.

I/O & resources

  • Definition of actor classes and their decision rules
  • Data on networks, resources and initial states
  • Assumptions on environmental dynamics and exogenous influences
  • Simulated time series and spatial patterns
  • Sensitivity and scenario comparisons
  • Empirical indicators for decision support

Description

Agent-Based Modeling (ABM) is a simulation-based method for studying complex systems by modeling autonomous agents and their interactions. It enables analysis of emergent phenomena, policy experiments, and scenario testing. ABM is used in social sciences, ecology, and economics to reveal micro–macro linkages.

✔

  • Enables exploration of emergent phenomena arising from local interactions.
  • Flexible for heterogeneous actor and network structures.
  • Well suited for scenario and policy testing under varying assumptions.

✖

  • High data requirements to calibrate realistic models.
  • Computational effort scales with agent count and interaction density.
  • Challenges in generalizing results across scenarios.

  • Emergence metrics

    Measures that quantify collective patterns and deviations from expected behavior.

  • Calibration error

    Difference between model outputs and observed data to assess model fit.

  • Runtime/scaling metrics

    Measurement of simulation runtimes depending on agent count and interaction density.

Schelling's segregation model

Classic ABM demonstrating how individual preferences can lead to segregated patterns.

Epidemiological ABM studies (e.g. COVID models)

Models that simulate infection dynamics and intervention effects in heterogeneous populations.

Traffic simulations with NetLogo and Mesa

Practical examples representing agent movements and congestion behavior in networks.

1

Define research question and metrics

2

Model agents, rules and environment

3

Implement in an ABM platform and run initial tests

4

Calibration, validation and sensitivity analysis

5

Perform scenario sweeps and document results

⚠️ Technical debt & bottlenecks

  • Spaghetti code from incremental extensions without refactoring.
  • Insufficient testing and validation infrastructure.
  • Missing automation for experiment reproducibility.
Compute resources for large agent countsData availability and quality for calibrationValidation and verification of complex models
  • Directly transferring local model results to other contexts without checks.
  • Using ABM solely for visualization without hypothesis testing.
  • Using unavailable or unsuitable data for calibration.
  • Underestimating nonlinear behavior and sensitivity effects.
  • Failure to separate model assumptions from empirical findings.
  • Overfitting by over-optimizing to a single observed dataset.
Knowledge of modeling and complexity theoryProgramming skills (e.g. Python, NetLogo or Java)Statistical analysis and calibration techniques
Heterogeneity of agentsScalability and compute performanceNetwork and interaction structures
  • • Limited compute capacity for real-time needs
  • • Missing or unreliable observational data
  • • Regulatory and data protection requirements