Developing an ABM-driven Decision Support System

in the Emergency Services

Stephen Dobson

, Mark Burkitt

, Dermot Breslin

and Daniela Romano

International Centre for Transformational Entrepreneurship, Coventry University, Coventry, U.K.

Department of Computing, University of Sheffield, Sheffield, U.K.

Management School, University of Sheffield, Sheffield, U.K.

The Department of Computing, Edge Hill University, Lancaster, U.K.

Keywords: Decision Support System, Co-evolution, Connectivity, Agent-Based Model, Domestic Fire Risk Behaviour.

Abstract: The research presented here outlines an application of Agent-Based Modelling (ABM) used to support

strategic decision-making in the emergency services. Here the resulting ABM-driven Decision Support

System (DSS) (called ‘Premonition’) is designed to help practitioners engage with a complex and dynamic

environment of co-evolving fire risk behaviours through time. Social change is presented here as a process by

which behaviours co-evolve within connected networks of agents. ABM is identified as a beneficial approach

to simulate changing household behaviours and the influence of social networks, environmental factors, and

also fire service interventions within the Sheffield City region, UK. This project seeks to develop a DSS which

supports the stages of ‘intelligence’, ‘design’, and ‘choice’ as the decision-maker moves from identifying

problem areas, establishing possible strategies for intervention, and predicting possible outcomes of

alternative courses of action.

1 INTRODUCTION

As public services face budgetary cuts in many

European countries, a ‘more-for-less’ mantra pushes

public sector managers to look for greater efficiencies

and value added decision-making. This is particularly

so with emergency services who are faced with

frontline realities on a daily basis. Strategic decision-

making in this arena aims to support the protection of

some of society’s most vulnerable, whilst balancing

decreasing budgets. Decision Support Systems (DSS)

may be seen as an ever more valuable part of the

effective management of resources in this public

service context. DSS have gained increasing

relevance in business and industry since the 1970s,

and particularly with the advent of data warehouses,

on-line analytical processing, data mining and the

Web in the 1990s. However, this paper focuses on

current developments in model-driven DSS as an

important support tool for managers to embrace a

more dynamic and evolving (co-evolving) problem

environment. The paper focusses on Agent-Based

Modelling (ABM) as an emerging approach in DSS

to help develop decision-makers’ understanding of

the complex social environments through which they

operate.

Understanding and predicting the behaviours of

households within a community is an important factor

in the planning and operationalisation of activities in

public service. For example, an important part of the

South Yorkshire Fire and Rescue, UK (SYFR)

Integrated Risk Management Plan 2013-17 has

involved the exploration of community fire risk.

Through the geo-demographic mapping of socio-

demographic lifestyle profile data (MOSAIC) the

service aims to improve their information provision

to the public and fire prevention work. Previous

academic studies can support our understanding of

community behaviours in relation to fire risk so as to

help services work on prevention measures in both a

strategic and targeted manner. For example, research

has shown that fire risk due to the behaviours of some

households differ for particular demographic and

socio-economic groups (Smith et al., 2008; Taylor et

al., 2012). By using socio-economic factors to

classify households into low, medium, and high fire

risk groups, preventative measures in the form of

information and support may be targeted at those

most in need. The expectation therefore is that

aggregate risk may be reduced over time in these

localities thus reducing the frequency of emergency

interventions in those geographic areas.

Dobson, S., Burkitt, M., Breslin, D. and Romano, D.

Developing an ABM-driven Decision Support System in the Emergency Services.

In Proceedings of the 18th International Conference on Enterprise Information Systems (ICEIS 2016) - Volume 2, pages 151-157

ISBN: 978-989-758-187-8

151

Secondly, statistical approaches may be used to

carry out detailed historical analyses of changing

patterns of behaviour. This enables resource planning

and allocation and for decision-makers to anticipate

need. This kind of method represents a simple model-

driven approach to DSS but it is not without

limitations. By classifying households into groups

based upon socio-demographic factors there is an

assumption that behaviours are static. In this sense,

membership is ascribed based on a notion of shared

characteristics and therefore fire risk by association is

intrinsic. The model does not account for changing

behaviours or the influence of one’s environment or

social connections. Using a linear model for

projecting historical data forward to predict future

outcomes will again assume a stable (static) problem

environment.

This paper outlines research into the development

of a practical model-driven DSS for implementation

in the emergency services with SYFR. It uses ABM

to develop a dynamic understanding of changing

behaviours within a community. Whilst the model

draws upon socio-demographic and historical data as

used in the approaches outlined above, the research

also explores the interacting mechanisms which drive

behaviour change. As such the authors present a

simple framework for modelling social change seen

as a process of co-evolution between community

networks (Breslin et al., 2015; Dobson et al., 2013).

This approach shifts the focus of attention from

individual households to the connected and co-

evolving sets of behaviours that they represent. To

further advance this approach, we have developed an

ABM-driven DSS to simulate changing household

behaviours within the Sheffield City region. This

DSS draws on historical data of fire incidents and

direct community interventions undertaken and

collated by SYFR over the last 5 years (e.g. home

safety checks, fire safety campaigns etc.). The project

aims to improve organisational decision-making

through the three stages outlined by Simon (1960).

These are: a) the 'intelligence’ stage of searching for

problems (i.e. the identification) by helping to

identify areas of risk; b) the ‘design’ of inventions

aimed at reducing risk; and c) ‘choice’ of a course of

action based upon predicted outcomes of

interventions (i.e. the exploration of alternatives). As

such it is anticipated that the developed DSS will

become a valuable tool in optimisation of resource

allocation planning of operations as well as

community prevention work.

2 DECISION SUPPORT SYSTEMS

Decision Support Systems (DSS) can be defined as

“computer-based tools that help users in a problem

solving environment to improve their productivity

and decision-making ability” (Bayraktar and Hastak

2009, p1357). Since the 1970s DSS have evolved and

gained greater prominence in organisational strategic

and operational decision-making (Lauria and

Duchessi, 2006; Chung et al 2004; Shim et al., 2002;

Keenan 1998; Jagielska, 1993; Jensen, 1990). Shim

et al (2002) outline that it was Gorry and Scott Morton

(1971), whose integration of Anthony’s (1965)

categories of management activity and Simon’s

(1960) description of decision types, were perhaps

most influential in defining the concept of DSS. For

Anthony (1965), management activities involved

strategic, management (or tactical), and operational

controls; layers of organisational decision-making

which are now universal tenets in management

thinking. Simon (1960) presented problems for

decision-makers as existing on a continuum from

‘programmed’, which are routine and repetitive

problems that are well structured and easily solved, to

‘non-programmed’. Non-programmed problems are

unique, ill-structured and more difficult to solve, i.e.

'wicked problems’ (Rittel and Webber 1973). To

represent this continuum Gorry and Scott used the

terms ‘structured’, ‘semi-structured’, ‘unstructured’

to develop a framework based upon Simon (1960)

incorporating the search for problems (intelligence),

the development of alternatives (design), and the

analysis of alternatives (choice) (Shim et al 2002).

DSS can be thought of as incorporating a wide

range of technics and applications to help decision-

makers with the processes of 'intelligence', 'design'

and 'choice' such as artificial intelligence, expert

systems, database querying and analytical predicative

modelling. However, the purpose of DSS is not to

provide a direct solution, but simply to add value to

any system output when reaching a decision

(Bayraktar and Hastak 2009).

From the early 1990s DSS benefitted from

emerging developments such as data warehouse, on-

line analytical processing (OLAP), data mining, and

the Web. Lauria and Duchessi (2006) outline that

DSS tends to fall into two main categories comprising

of either narrow or broad definitions. The narrower

or data-driven definitions include these emerging

applications, such as OLAP and data mining, which

enabled users to combine numerous databases

through explicitly defined ontological relationships

so as to enable multiple combinations of data

elements for analysis over time. Data was able to be

ICEIS 2016 - 18th International Conference on Enterprise Information Systems

152

presented in various graphical formats to support

decision making. “Data mining applications identify

specific, unknown patterns in databases and data

warehouses that typical queries cannot reveal” (Ibid

2006, p1574).

Broader, model-driven DSS aimed to expand

upon techniques to solve complex and sometimes

unstructured problems using quantitative and/or

qualitative models. A model-driven DSS comprises

of three components (Shim et al., 2002) which

integrate, 1) data management/database; 2) model

management involving one or more models applied to

the problem, and; 3) dialogue management which

enables users to change some input variables and

initiate database and model management elements of

the DSS: “via the dialog module, users interact with a

DSS and can perform sensitivity, or “what-if,”

analyses to gain more insight into the problem and its

potential solutions” (Lauria and Duchessi, 2006,

p.1574).

Numerous model-driven methods of DSS have

supported the management of risk in decision-making

such as Decision Tree Analysis (Apolloni 1998),

Cause-Consequence Analysis (De Meaux and

Koornneef 2008; Lee et al 2008), Analytic Hierarchy

Process (Andreica 2009), Monte Carlo method (Hart

2008; Horstmann 2006) and Bayesian Networks

(Xiacong and Ling 2010). The modelling which

supports DSS has been subject to much research and

Cho (2007) identifies three generic strategies which

have shaped development in organisational and

decision-making arenas.

Initially reductive and simplistic computational

models have aimed to isolate key causal relationships

between data (Dodin and Elmaghraby 1985;

MacCrimmon and Ryavec 1964). To address the

limitations of these approaches, scholars have

pursued a second, more computationally intensive,

means of increasing predictive accuracy. Cho (2007)

identifies broad approaches (which include some of

those described above) within this second strategy,

these are the one-time update approach; the Markov

Chain Monte Carlo (Virto et al 2002), and Bayesian

networks. The Bayesian approach (Covaliu and

Soyer 1996) involves developing an acyclic network

of sequential and causal activities or elements in a

model with assigned probabilities. As elements

become known through observation within the

network, probabilities of outcomes are updated

according to Bayesian statistical inference. A third

strategy lies between these two ultimately static

approaches, which aims to maintain both levels of

simplicity whilst also accounting for complex

cyclical and dynamic dependencies and interactions

of feedback between activities.

DSS in the 21

century is described by scholars as

being characterised by collaborative decision-making

and collaboration platforms to support increased

connectivity and data sharing. Here we see a

migration of decisions made by individuals to ones

made by diverse groups or even multiple firms (Shim

et al 2002). However, as Keen (1987) identified in

the late 1980s, there is also a need to explore increases

in computing power to extend model-driven DSS to

embrace much more fluid, emergent and nuanced

picture of the decision environment. Mitroff and

Linstone (1993) suggest that this kind of shift would

include consideration of much broader organizational

and cultural factors than have featured in past DSS.

In the work presented here we present DSS in the

public service environment as needing to reflect

changing and dynamic processes of co-evolving parts

within a social system. The emergency services is

used here as an example of how a more responsive

and dynamic decision-making context can benefit

from an ABM-driven DSS approach.

3 AGENT BASED MODELLING

“Agent-based modelling is a computational method

that enables a researcher to create, analyse, and

experiment with models composed of agents that

interact within an environment”. (Gilbert 2008, p2)

Agent-based models (ABM) consist of agents

modelled to interact with each other and the

environment through a set of predefined rules or

heuristics. Agents are broadly defined and are

distinct parts of a program representing social entities

such as individuals, groups, organisations or wider

social, political or economic institutions. Agents may

be defined as interacting and responding to physical

feedback through movement, as explored through the

analysis of swarm behaviour, or as in the case of this

research may be purely social entities interacting on a

behavioural level. As such, broad ranges of complex

adaptive systems have been modelled using ABM

frameworks. A Flexible Large-scale Agent

Modelling Environment on the Graphics Processing

Unit (FLAME GPU) (Richmond et al 2010) has been

used for this project. The technology merges the

modelling power of ABM with that of 3D graphics.

The resulting framework enables very large-scale

simulations with a massive number of agents to be

processed and visualised in real-time with both 2D

and 3D representation of the environment. Examples

Developing an ABM-driven Decision Support System in the Emergency Services

153

of FLAME GPU ABM research range from

modelling the illegal drug market (Romano et el

2009), to investigating innovative methods for

training public service staff working within the

community, to a simulated social crowd for the

training of CCTV operators to spot malicious

behaviours.

3.1 ABM-driven DSS

Given the complex and dynamic nature of decision-

making environment through which SYFR operate

(due to shifting household behaviours over time) it is

suggested here that the co-evolutionary process

through which change occurs may be further

understood through models. As described above,

previous approaches have used models to estimate

fire risk which have remained statically defined and

therefore unresponsive to change. By extending these

methods through the development of an ABM-driven

DSS we propose that the dynamics of change are

better reflected to the decision-maker. In ABM the

changing behaviours of specific households (or

‘agents’) are modelled computationally. These agents

are viewed as interacting heuristically within a

network of other households which define the

geographic community. The heuristics and agents for

the model are broadly outlined in the next section, and

also underpinned by literature reported on more fully

in a forthcoming publication dedicated solely to the

model specification. These include the influence of

connections upon household behaviour within a close

social network in relation to changing domestic

behaviours (i.e. smoking, consumption of alcohol,

use of electrical appliances and cooking practices). A

target community in South Yorkshire (United

Kingdom) was simulated over time through an ABM

approach to illustrate the emergence of patterns of

behaviour within a complex system of interacting

parts. A key advantage of ABM for DSS is that it

provides a simulated dynamic environment enabling

the decision-maker to carry out ‘what-if’ experiments

which would otherwise be impossible in live

scenarios. This helps the decision-maker to develop

approaches and design choices which maximise

opportunities for positive change.

4 THE MODEL

The purpose of the model is to investigate the impact

that different intervention methods are likely to have

on fire risk in different areas, based on knowledge

about the people who live in the area. Each individual

household in the area is explicitly modelled, and

includes details about the individual behaviours and

fire risk factors of the people living in the household,

as well as a representation of their household level

social network. The simulation represents a set of

different intervention methods, which can be used to

predict which interventions are likely to have the

most impact under different conditions. Each

intervention can have both a direct influence on a

household, and an indirect influence via the

household’s social network.

Data on fire service callouts has been provided for

the development of the DSS ranging from 1st April

2009 until 1st December 2014. This contains

deliberate, accidental and unknown incidents, both

dwelling and non-dwelling. Each record contains

detailed information about the incident, along with an

Ordnance Survey grid reference.

Data from the Home Safety Check (HSC) reports

has also been provided for the period between 2009

and 2015. Each record in the data set contains a street

address, date, a single question and corresponding

answer. There are 70103 data rows, which

correspond to roughly 1589 households, which have

one or more repeat checks for different years.

Finally, MOSAIC demographic data containing

household level classifications for every household in

the area was also used to create the agent-based model

of fire-risk behaviours and comprised of 4723 unique

records in total for the area.

4.1 State Variables and Scales

Three different agent types have been identified;

these agents represent an individual household, fire

incidents and interventions.

The Household agent represents an individual

household within the area. This can be the occupants

of an individual house, or a flat within a larger

building. The Household Agent contains all the

parameters that define an individual household in the

area. Each household has a unique identifier and a

location which determines its physical location in

relation to other households. A household also has a

set of Risk Markers, which are used to determine the

Risk Factor, which is the relative risk that an

individual household will have a fire. The Risk

Markers are derived using all available information

about the households in the area. The Risk Factor

fluctuates in response to external influences and the

gradually decays over time back to the baseline.

As individual households are influenced by both

interventions and the influence of social connections,

the Household agent maintains an individual social

ICEIS 2016 - 18th International Conference on Enterprise Information Systems

154

network, which connects it to other households in the

area. These relationships are modelled as properties

of a Household, each with a directionality and

weighting, representing the amount of influence that

one household has over another.

The households are influenced both directly and

indirectly via their relationships by interventions and

fire incidents. Intervention agents can have different

types, representing the range of different

interventions that can be performed, such as leafleting

campaigns, home safety visits and shock campaigns.

An intervention will typically have an area of

influence, a strength, representing its effectiveness,

and duration, representing how long the influence of

the intervention will last. A FireIncident agent

represents an individual household fire, which could

potentially spread to other households, and cause

fatalities in one or more households. For historical

time periods, fires are based on the actual fire

incidents. For future time periods, fires are random,

based on statistical analysis of the available data.

When a fire occurs, it is assumed that the occupants

won’t change. However, there will be an initial area

effect of reduced risk in the surrounding households.

The magnitude and radius of the effect will be related

to the incident severity. Changes to the

neighbourhood are not modelled, and the households

remain the same for the duration of the simulation.

A general systems diagram illustrating the DSS

(called ‘Premonition’) can be seen in figure1. Here

the inputs may be regarded as both data and also

management decisions. Management decisions

comprise of the levels suggested by Anthony (1965).

These are operational (planned operational

interventions and infrastructure), tactical

(performance measures) and strategic (policy and

budgetary considerations). The ‘Premonition’ ABM-

driven DSS combines these data and organisational

knowledge inputs through a model management and

processing component and dialogue management

(graphical user interface, GUI). The key output is in

the form of GIS mapping presented sequentially with

Figure 1: General system diagram for the ‘Premonition’ ABM-driven DSS.

DSS

INPUT

Historical

Incidents

MOSAIC

demographics

Historical

Interventions

Social network

influence

Data

Performance

measures

Policy /

budgets

Operations and

infrastructure

Management Decisions

GUI

OUTPUT

Visual GIS

mapping

Household risk

rating

Area risk rating

Statistical

reports

Incident

influence

Intervention

influence

Projections

Literature

Developing an ABM-driven Decision Support System in the Emergency Services

155

1 day time steps illustrating incidents, interventions

and shifting household fire risk over time.

5 DISCUSSION

This work adds to ongoing research which considers

the dynamics of networking processes and the nature

of connected behaviours which change and develop

over time. As such the resulting model aims to

‘embed’ the decision-maker in this process to provide

a richer and more nuanced picture of the ever

changing environment that they are working in. The

more cohesive and close-knit the network, the more

interpretive heuristics are shared between households

(Breslin, 2011; Dobson et al., 2013), and as a result

the fewer the opportunities for different

interpretations and with this possibilities for variation

and innovation. In these close-knit communities,

local authorities and service providers need to be

closely engaged and embedded in order to affect

change. Interventions should therefore be targeted at

key thought leaders, and positioned in terms of local

issues. On the other hand, in more sparsely connected

networks opportunities for change are increased.

However, interventions here would tend to be more

costly given the difficulty in reaching such diverse

groups. In both instances households which are more

socially isolated present the greatest challenge.

Validation is key to developing the predictive

power of ABM. In a sense the validation process

involves comparing and then fine-tuning the model to

reflect actual recorded behaviours of households. In

this study data will be drawn from three key sources.

In the first instance, a generic model of changing fire

risk behaviours is constructed drawing from

extensive academic and industry-focused research

noted above. Second, past research on fire risk

behaviours and trends, including both published

reports and data obtained by SYFR, is used to adapt

the generic model. Finally, region-specific data

collated by SYFR is used to fine tune the model. This

final process allows both historic data trends and the

local experience and expertise of SYFR staff to be

incorporated into the model. Once validated, the

model can be expanded to include other regions both

locally and nationally, to predict changing trends in

fire risk behaviours. In addition the model can be

further developed to consider other types of

household behaviours, for example related to health

and social care. In sum, once validated, the co-

evolutionary model can be generalised to simulate

changes in other household behaviours using the

developed understanding on connected social

networks.

6 CONCLUSIONS

Understanding and predicting the behaviours of

households within a community is an important factor

in the planning and operationalisation of activities for

the SYFR emergency service. To help support

decision-makers in this environment the paper has

focussed on the development of an ABM-driven DSS

to help develop decision-makers’ understanding of

the complex social environments through which they

operate. Whilst the model has drawn upon socio-

demographic and historical data as used in existing

approaches, the research has also explored the

interacting mechanisms which drive behaviour

change. The framework for modelling social change

is seen as an ever shifting interaction between agents

within a system. Three key agent types have been

identified in this study representing individual

households, fire incidents, and interventions. The

DSS is able to provide decision-makers with an

historical view of fire events and interventions in the

target area but also model changes in fire risk

behaviours. Whilst socio-demographics play an

important part in the baseline risk of households,

interventions, fire events and the influence of primary

and secondary social networks all combine to

influence changing fire risk behaviours over time.

Using this tool it is anticipated that decision-makers

may calculate the modelled aggregate risk for an area

and explore ‘what-if’ scenarios of various possible

planned interventions such as targeted information

provision and community support.

ACKNOWLEDGEMENTS

The authors would like to thank Steve Chu, Steve

Fletcher, Graham Howe, Mick Mason, Nicola Smith,

Ian Standeven and colleagues at South Yorkshire Fire

& Rescue for help and support in the development of

this research.

REFERENCES

Andreica, M. 2009. ‘Overview of Modelling the Risk

Assessment Process in Applied Project Management’

Metalurgia International, 14, pp. 139-144.

ICEIS 2016 - 18th International Conference on Enterprise Information Systems

156

Anthony, R.N. 1965. Planning and Control Systems: A

framework for analysis Harvard University Graduate

School of Business Administration

Apolloni, B. 1998. ‘Learning fuzzy decision trees’ Neural

Networks, vol. 11, pp 885-895.

Bayraktar, M.E. and Hastak, H. 2009. ‘Bayesian Belief

Network Model for Decision Making in Highway

Maintenance: Case Studies’ Journal of Construction

Engineering and Management 135(12), pp.1357-1369

Breslin, D. 2011. Interpreting Futures through the Multi-

Level Co-Evolution of Organizational Practices.

Futures, 43(9), pp.1020–1028.

Breslin, D., Romano, D. and Percival, J. 2015.

‘Conceptualizing and Modelling Multi-Level

Organisational Co-Evolution.’ In Secchi, D. and

Neumann, M. (Eds.) Agent-Based Modeling in

Management and Organizations, Springer, pp. 137-157

Chung, T. H., Abraham, D. M., and Gokhale, S. B. 2004.

‘Decision support system for microtunneling

applications.’ Journal of Construction Engineering

Management, 130(6), pp. 835–843.

Covaliu, Z., and Soyer, R., 1996. ‘Bayesian project

management’. In Proc. Conf. ASA section on Bayesian

Statistical Science, pp. 208–213.

de Meaux, J. and Koornneef, M. 2008. ‘The cause and

consequences of natural variation: the genome era takes

off!’ Current Opinion in Plant Biology, 11, pp. 99-102.

Dobson, S., Breslin, D., Suckley, L., Barton, R. and

Rodriguez, L. 2013. Small Firm Growth and

Innovation: an Evolutionary Approach. International

Journal of Entrepreneurship & Innovation, 14(2), 69-80.

Dodin, B.M., and Elmaghraby, S.E., 1985. ‘Approximating

the criticality indices of the activities in PERT

networks’. Management Science 31, pp. 207–223.

Gilbert, G. N. 2008. Agent-based models Sage Publications.

Gorry, G.A. and Scott Morton, M.S. 1971. ‘A Framework

for Management Information Systems’ Sloan

Management Review 13(1), pp50-70

Hart, D. 2008 ‘Risk management in paediatric hospitals:

Results of the multicentre project “Reducing Risks -

Increasing Safety"’. Monatsschrift Kinderheilkunde,

156, pp. 1104-1113.

Horstmann, R. 2006 ‘Risk management in the operation

room results of a pilot project of interdisciplinary

"incident reporting", Zentralblatt Fur Chirurgie,. 131,

pp. 332-340.

Jagielska, A. and Jacob, 1993. ‘A Neural network model for

sales forecasting’. New Zealand International

Conference on Artificial Neural Networks and Expert

Systems Dunedin, New Zealand, pp. 544–549

Jensen, F., Lauritzen, S. and Olesen, K. 1990. ‘Bayesian

updating in causal probabilistic networks by local

computations’ Computational Statistics Quarterly 4,

pp.269–282.

Keen, P. 1987. ‘Decision Support Systems: The next

decade’ Decision Support Systems 3(3), pp253-265

Keenan, P. 1998. ‘Spatial Decision Support Systems for

vehicle routing’, Decision Support Systems 22(1) pp.

65–71.

Lauría, E. J., and Duchessi, P. J. 2006. ‘A Bayesian belief

network for IT implementation decision support’.

Decision Support Systems, 42(3), pp. 1573-1588.

Lee, E., Park, Y. and Shin, J.G. 2008 ‘Large engineering

project risk management using a Bayesian belief

network’ Expert Systems with Applications 36, pp.

5880–5887

MacCrimmon, K.R., and Ryavec, C.A., 1964. ‘An analytic

study of the PERT assumptions’. Operations Research

14, pp.16–37.

Mitroff, I.I., and Linstone, H.A. 1993 The Unbounded

Mind: Breaking the chains of traditional business

thinking Oxford University Press, New York

Richmond, P., Walker, D., Coakley, S., & Romano, D.

2010. ‘High performance cellular level agent-based

simulation with FLAME for the GPU’.Briefings in

bioinformatics, 11(3), pp.334-347.

Rittel, H.W., and Webber, M.M. 1973. ‘Dilemmas in a

general theory of planning’. Policy sciences, 4(2), pp.

155-169.

Romano, D. M., Lomax, L., & Richmond, P. 2009.

‘NARCSim an agent-based illegal drug market

simulation’. In Games Innovations Conference, 2009.

ICE-GIC 2009. International IEEE Consumer

Electronics Society's (pp. 101-108). IEEE.

Shim, J., Warkentin, M., Courtney, J., Power, D., Sharda,

R. and Carlsson, C. 2002 ‘Past, present, and future of

decision support technology’, Decision Support

Systems 33 pp. 111–126.

Simon, H.A. (1960) The New Science of Management

Harper Brothers, New York

Smith, R., Wright, M., and Solanki, A. 2008. Analysis of

fire and rescue service performance and outcomes with

reference to population socio-demographics.

Department for Communities and Local Government,

Fire Research Series 9/2008

Taylor, M.J., Higgins, E., Lisboa, P.J., and Kwasnica, V.

2012. ‘An exploration of causal factors in unintentional

dwelling fires’. Risk Management, 14, pp. 109 – 125

Virto, M.A., Martin, J., and Insua, D.R., 2002. ‘An

approximate solutions of complex influence diagrams

through MCMC methods’. In: Gamez, S. (Eds.), Proc.

First European Workshop on Probabilistic Graphical

Models, pp. 169–175

Xiaocong, H., & Ling, K. 2010. ‘A risk management

decision support system for project management based

on Bayesian network’ in Information Management and

Engineering (ICIME), 2010 The 2nd IEEE

International Conference, pp. 308-312.

Developing an ABM-driven Decision Support System in the Emergency Services

157