Agent-Based Models to Predict Crime at Places

Overview

• Agent-based modelling: It is a methodology used in computer simulation to model crime.
  • It is ideal for crimes like as burglary and street crime, which are significantly influenced by ambient conditions and individual behavior.
  • Agents are placed in a setting that permits them to travel through space and time and behave as they would in the real world.
• ABM is also a computer simulation technique that focuses on individual-level behaviour and is perfectly suited for crime modelling.
  • It entails placing virtual "agents" in an environment that lets them to traverse space and time and behave as they would in the real world.
  • ABM can be used both to investigate Agent-Based Models to Predict Crime at Locations and to produce accurate predictions in a virtual environment that simulates the actual world.
  • This post examines agent-based modeling and the benefits it provides to the subject of environmental criminology.

Crime Is a Complex System

• Social systems: These are complex systems comprised of numerous components, intricate interactions between those components, and emergent traits.
  • Acquisitive crimes: Such as burglary and robbery on the street, are an illustration of how environmental, social, and individual forces mix to produce a complicated collection of processes.
  • Understanding the processes and forces that define the criminal justice system is essential for crime prevention and policy formulation.
  • Contemporary criminology emphasizes the significance of "microplaces" that serve as the setting for a particular crime, and research conducted at scales greater than houses and streets obscures crucial criminal dynamics.
  • The same scale concerns apply to individuals participating in a crime occurrence, as aggregate treatment of individuals is likely to overlook the crucial everyday dynamics of individuals.
• Agent-based modeling is utilized to transition away from aggregate models and toward individual-level models.
  • It consists of autonomous entities known as "agents" that are capable of making decisions and interacting with each other and their surroundings.
  • This method enables the incorporation of genuine human behavior into computational models, such as crime simulations.
  • Creating a model from the bottom up is a more natural way to represent a complicated system than developing rules at an aggregate level.

A Description of Agent-Based Modelling

• ABM is a new system modeling technique that comprises of virtual "agents" capable of autonomous behavior in a virtual environment.

  • It is especially relevant to criminology since it permits the use of models to conduct studies that would otherwise be difficult or immoral.

• There are numerous definitions of "agent," however from the perspective of crime modelling, the following are routinely applied:

  • Autonomy: An agent should be able to regulate its own state, engage with other agents and its environment, and make choices without being directly governed by a central authority.
  • This appears to be an appropriate method for modeling people, such as offenders, victims, and other required parties (parents, managers, etc.).
  • Heterogeneity: Agents don’t need to be identical.
  • It is possible to develop offender agents that reflect the variety of offending behaviors demonstrated, allowing for the incorporation of qualitatively collected facts and theories.
  • Reactivity: Agents should be able to respond to environmental changes in a proactive manner, demonstrating goal-directed behavior.
  • This is especially relevant for a crime model, as the environment will alter as a result of the crime, influencing the agents' future conduct.
  • Bounded rationality: It is essential, especially in social science modeling, that agents do not always act exactly logically.
  • Agents can be designed with "bounded" rationality by restricting their world knowledge, so that their decisions are not always optimal.

• ABM provides a natural explanation of complicated systems, which mathematical formulae cannot typically describe.

• To comprehend geographical human systems, it is vital to comprehend the thinking behind individual decisions, and it is more natural to represent individuals directly than to attempt to manage them through aggregate equations.

• The manually built model of home stability by Schelling demonstrates the usefulness of ABM for modeling social systems.

  • It consists of a one-dimensional habitat filled by homes from one of two racial groups, with a global parameter dictating the proportion of the same race that each household want to reside close to.
  • The agent-based modeling software "NetLogo." includes a two-dimensional version of this as an example.

• Schelling's findings indicate that a preference to live adjacent to no more than 50 percent of the same racial group results in blatant segregation.

  • This indicates that high segregation can emerge from individuals with moderate segregation preferences.

• Agent-based models (ABM) are a more naturalistic method of modeling a system since they describe the system's basic components and then attempt to "grow" the observed crime patterns from the "bottom up"

• ABM is particularly useful in environmental criminology since offenders can be treated similarly to non-offenders and the effects of noncriminal activities on crime can be investigated.

• However, ABM has some disadvantages that must be addressed. Research on crime modelling is tough due to the difficulties of modeling "soft factors" such as irrational human behavior and complex psychology.

  • This is exacerbated by the limited availability and imprecision of crime statistics.

• There are also significant challenges associated with the model's implementation, such as requiring a high level of computer programming expertise and being processor- and storage-intensive.

• With some of the most modern and sophisticated techniques, ABM is becoming an increasingly popular tool for crime analysis.

State of the Art

Offender Behavior

• There are cognitive architectures that can be utilized to model human behavior, despite the fact that agent-based criminal modelling is an incredibly difficult task.
• Malleson (2010) employ the PECS model of human behavior to regulate agents by comparing the magnitudes of various motivations.
• By altering the manner in which different motivations influence the agents' behavior, it is feasible to generate distinct sorts of criminal agents that represent contemporary criminological thought.
• By placing virtual offenders in a setting that closely resembles that of the real world, the authors are able to investigate the potential real-world crime patterns that may occur under various scenario situations.

Street Networks

• Groff (2007) examined the applicability of routine activity theory to street robbery using agent-based modeling.
• Citizens (offenders, victims, and guardians) and police make up the model's two agent kinds.
• Citizen agents are assigned a home location at random and spend their time away from home visiting randomly assigned employment and activity nodes.
• The decision to commit an offense is determined on the level of guardianship and the wealth of the possible victim at the offender's current location.
• The model indicated that the number of street robberies grew when residents spent more time away from home and had more opportunities to encounter potential perpetrators.
• Even if the travel patterns of the agents were random, certain street crossings revealed substantial event clusters.
• This provides insight into how the urban structure of streets might affect the locations of street crime, using actual city streets as examples.

Social Networks and Cohesion

• Hayslett-McCall et al. (2008) include measures of social cohesiveness and guardianship in their household burglary model, which assigns attractiveness scores depending on socioeconomic position, guardianship, race, and economic levels.
• Dray et al. (2008) integrate linkages between local actors, authorities, social services, and drug users and sellers in an ABM of the drug market in Melbourne.
• With this paradigm, small-scale efforts disrupted drug markets more effectively than national ones.

Abstract Theory

• The models presented thus far situate crime within an actual environmental context by utilizing realistic virtual settings.
• Yet, precisely anticipating spatiotemporal crime trends need not be the focus of the investigation.
• Brantingham (2004) have created a criminal model in which agents can move across time and space and interact with one another and the environment.
• This model can serve as a "virtual laboratory" to investigate the individual-level dynamics that occur as a result of various crime theories in an environment that lacks the complexity of the real world.
• The resulting simulation, A 46 Agent-Based Models to Forecast Crime at Locations, can be used as an inter-disciplinary instrument to aid criminologists in examining the dynamics of urban crime.

Controversies in the Literature

• Schelling's segregation model is the more abstract of the two types of agent-based models: pessimistically abstract or optimistically realistic.
• Modellers concentrate on developing simple models based on restricted subsets of rules and environmental conditions.
• Elffers and van Baal (2008) argue that simple models can be a potent explanatory force, but that patterns can match models even if their explanations are different.
  • It is challenging to verify that human systems would behave similarly to these simplistic models.
• Groff (2007) and Malleson et al. (2010) coupled their models with a geographical information system (GIS) to tackle the identifiability issue and produce more trustworthy comparisons with reality.
• However, the enormous number of variables may cause the models to be overly adaptable, which raises the issue of identifiability.
• It is frequently challenging to verify crime-study models with new data, and it is uncertain how well the data reflects actual crime patterns.
• Models are best constructed in a step-by-step manner, first ensuring that simple behaviors with simple system components function as predicted, and then increasing their complexity until it more closely reflects reality.