Abstract Behavioral Modeling or ABM is an approach to model social and natural systems using the behaviors of its individual components. ABM is a class of modeling techniques where individual agents are defined by their behaviors and attributes. These agents interact with each other and their environment resulting in emergent behaviors at the system level. ABM is advantageous as it can capture emergent phenomena that may not be predicted by other simpler mathematical models. This paper reviews the fundamentals of ABM and demonstrates a sample model using the NetLogo modeling platform to simulate viral spread during a pandemic.
Introduction
Agent-Based Modeling (ABM) is a computational modeling methodology that enables modeling of complex systems by explicitly representing individual components called agents and their interactions. The behavior of agents and their interactions give rise to the observed system level behavior which may not be predicted by analyzing the components individually. ABM’s roots can be traced back to ideas from complex systems, game theory and cellular automata. Some early applications included modeling of traffic flow, crowd dynamics and the spread of diseases.
ABM Concepts
The key elements involved in ABM are:
Agents – The basic components of the system. They represent individuals or collective entities that schedule and carry out activities depending on their attributes.
Environment – The space or domain in which agents exist and interact with each other directly or indirectly. It can be physical space like cities or ideological space like financial markets.
State variables – Attributes that define the internal state of an agent like location, age, energy level etc. These change over time based on agent behaviors and interactions.
Interactions – Rules that govern how agents interact and influence each other. Could be collision avoidance rules in crowd modeling or infection transmission in disease modeling.
Heterogeneity – Agents can have different attributes and follow different behaviors based on their type or internal state. This represents diversity in real world systems.
Emergence – Macro patterns or behaviors that arise from the local interactions between agents but cannot be explained or predicted from analyzing individual components.
A typical ABM consists of the following steps:
Identifying key components/agents in the system and their attributes. Some examples include individuals/households, resources, infrastructure etc.
Defining the rules and behaviors that govern how agents interact and how their attributes change over time based on these rules. For example, individuals may infect each other based on random encounters or travel patterns.
Setting up the environment/space for the agents and initializing their starting attributes.
Repeatedly executing the model by advancing agents through multiple time steps based on the defined behaviors and interactions. This helps simulate the evolution of the system over time.
Analyzing emergent patterns/properties at the system level that result from agent behaviors. Key metrics like rate of infection spread in a disease model are computed.
Comparing model output to real world data and validating results. Model parameters can then be adjusted iteratively to improve fit with empirical observations.
Once validated, the model can be used to conduct what-if analysis by changing underlying parameters to help understand dynamics, make predictions and evaluate counterfactual policies/scenarios.
ABM can capture features not exhibited by simpler modeling approaches:
Complex feedback processes where agent behaviors change other agents over time
Discrete decision making instead of assuming continuous optimization
Heterogeneity among components
Aggregation leading to macro patterns that are challenging to derive analytically
Complex spatial patterns especially in social and infrastructure networks
The limitations of ABM include its computational intensity, difficulty in validation, non-uniqueness of models for a given real system and limitations imposed by available empirical data to inform model parameters. Despite these challenges, ABM has received growing interest across diverse domains for studying emergent phenomena.
Sample ABM – Disease Spread Model
We present a sample ABM to simulate the spread of disease during a pandemic built using NetLogo, a popular platform for developing ABMs. The key aspects are:
Agents – Individuals residing in locations across a city. They have attributes like current location, infection status, age etc. Locations have attributes like maximum capacity.
Environment – Divided into a grid representing city blocks. Each cell of the grid is a location that can hold multiple agents.
Interactions – At each time step, a random proportion of agents move to adjacent locations based on daily routines. Infected agents have a chance to infect healthy agents at the same location based on an infection probability.
Heterogeneity – Agents have different daily routine probabilities based on their age which affects movement patterns. Infection risk depends on age.
Metrics – Monitor total infected and recovered agents over time to analyze disease spread dynamics.
The model is initialized by placing agents randomly across the grid respecting location capacities. A few random agents are designated as initial infected cases. The full code and user interface for running experiments are omitted here due to space constraints but sample outputs are shown:
https://i.imgur.com/FkeGOnu.png
This plots the number of infected agents over time for different values of the base infection probability. We can see higher probabilities lead to sharper rises in cases consistent with real pandemics.
ABM enables studying emergent effects of policies like lockdowns, vaccines, social distancing etc. For example, restrictions on travel between high infection locations slows disease spread. The impact of agent and location heterogeneities on transmission can also be captured. Overall, it provides a useful framework for building computational epidemiology models and analyzing non-pharmaceutical interventions during health crises.
Conclusion
This paper provided an overview of the ABM methodology including key concepts, model building process and demonstrated a sample disease spread model. ABM is a powerful tool to model emergent behaviors arising from self-organization in socio-technical systems. Its ability to represent heterogeneity, spatial interactions and feedback loops makes it suitable for epidemiology, social networks, transportation and many urban simulation problems. Future work includes model validation, sensitivity analysis, improving behavior representation using real data and exploring model capabilities to advise pandemic mitigation strategies. ABM offers a promising paradigm to study complex systems.
Andrew Stroud