GENT-BASED INTERDISCIPLINARY FRAMEWORK FOR

DECISION MAKING IN COMPLEX SYSTEMS

Marina V. Sokolova

Kursk State Technical University, Russia

University of Castilla-La Mancha, Department of Computing Systems, Spain

Instituto de Investigaci

on en Inform

atica de Albacete, Spain

Antonio Fern

andez-Caballero, Francisco J. G

omez

University of Castilla-La Mancha, Department of Computing Systems, Spain

Instituto de Investigaci

on en Inform

atica de Albacete, Spain

Keywords:

Decision support system, Agents, Complex system, Framework.

Abstract:

We offer a framework for the creation of decision support and expert systems for complex natural domains.

This is due, on the one hand, to the numerous advantages of intelligent methods of data manipulation and,

on the other hand, to the abilities of the computational agents to make decentralized decisions, which are

crucial for complex systems modeling and simulation.In our approach, the qualitative improvement in decision

making is obtained by using computational agents and interdisciplinary approach. The frameworks combines,

on the one hand, the numerous advantages of intelligent methods for data manipulation and, on the other hand,

the abilities of the computational agents to make decentralized decisions, which is crucial for complex system

modeling and simulation. The approach contributes to decentralization and local decision making within the

standard workﬂow. We demonstrate our framework in a case study and discuss obtained results.

1 INTRODUCTION

When the idea of creating a symbiotic human-

computer system to increment accessible knowledge

for decision making in complex problems appeared,

it was ﬁrstly applied for managerial and business do-

mains. Later, the initial domains of decision sup-

port systems (DSS) application were widened, and the

concept of DSS was spread out to manifold spheres

and ﬁelds of human activities, extending not only to

technical but also to complex ill-determined domains

(environmental, medical, social issues, etc). With

time diverse DSS have appeared, which enhanced a

number of models for use such as preprocessing, op-

timization, hybrid and simulation models.

The use of DSS is of great importance now for

complex natural phenomena studies, because they

allow specialists to quickly gather information and

analyze it in order to understand the real nature

of the processes, their internal and external de-

pendencies, and the possible outcomes while mak-

ing actions and correcting decisions. The tech-

nical areas where these systems could help vary

from the storing and retrieving of necessary records

and key factors, examination of real-time data gath-

ered from sensors, analysis of tendencies of com-

plex natural processes, retrospective time series, mak-

ing short and long-term forecasting, and in many

other cases (Ria

no et al., 2001),(Athanasiadis and

Mitkas, 2004),(Sokolova and Fern

andez-Caballero,

2008), (Sokolova and Fern

andez-Caballero, 2009).

In this article we will present our framework for

creation of agent-based decision support and expert

systems, focusing ﬁrstly on its principal formalisms

and phases of decision making process for a com-

plex natural domain (part 2) and revise related works

(part 3), then presenting our approach for creation a

framework for complex system design in complex do-

mains (part 4) and demonstrating an application of the

framework for the case study and discuss obtained re-

sults (part 5) and, lastly, we will comment directions

of future work (part6).

V. Sokolova M., Fernández-Caballero A. and J. Gómez F. (2010).

AGENT-BASED INTERDISCIPLINARY FRAMEWORK FOR DECISION MAKING IN COMPLEX SYSTEMS.

In Proceedings of the 2nd International Conference on Agents and Artiﬁcial Intelligence - Agents, pages 96-103

DOI: 10.5220/0002732000960103

 SciTePress

2 DECISION MAKING FOR

COMPLEX SYSTEMS

Human activity increases constantly, and both the

scale and speed of the human inﬂuence upon the nat-

ural, social, economical, and other processes grows

signiﬁcantly, so, it is impossible now not to take it

into account as one of the motive forces in ”human -

nature - technology” arena.

The science of today has reached signiﬁcant re-

sults in modeling and control over man-made tech-

nical systems. Notwithstanding, effective managing

natural complex phenomena often lies beyond of our

possibilities.

Generally speaking, a complex systems (CS) is a

composite object, which consists of many heteroge-

neous (and on many occasions complex as well) sub-

systems, and has emergent features, which arise from

interactions within the different levels. Such systems

behave in non-trivial ways, originated in composite

functional internal ﬂows and structures of the CS. As

a rule, researchers encounter difﬁculties when trying

to model, simulate and control complex systems.

Due to this, it would be correct to say that to puz-

zle out the natural complex systems paradigm is one

of the crucial issues of modern science. Because of

high complexity of CS, traditional approaches fail in

developing theories and formalisms for their analy-

sis. Such study can only be realized by a cross-

sectoral approach, which uses knowledge and theo-

retical backgrounds from various disciplines as well

as collaborative efforts of research groups and inter-

ested institutions.

The term ”system” has a number of deﬁnitions and

one of them sounds like ”a set of interdependent or

temporally interacting parts”, where parts are gener-

ally systems themselves, and are composed of other

parts in their turn (Levin, 2006). If any part is being

extracted from the system, it looses its particular char-

acteristics (emergency), and converts into an array of

components or units.

Thus, an effective approach to CS study is to fol-

low the principles of the system analysis (Rechtin,

1999), when we have to switch over to the abstract

view of the system and perform the following ﬂow of

tasks:

• Description of a system. Identiﬁcation of its main

properties and parameters.

• Study of interconnections amongst parts of the

system.

• Study of external system interactions with the en-

vironment and with other systems, etc.

• System decomposition and partitioning.

• Study of each subsystem or system part.

• Integration of results obtained on the previous

stages.

However, it is obvious that it is impossible to cre-

ate an overall tool, which would play a role of a

”magic wand” to study any complex adaptive prob-

lem. The process of DSS and expert system (ES)

design is laborious, and has many requirements, the

crucial one are: collaboration of the specialists from

various domains (experts, developers, analysts, math-

ematics, programmers, etc.). The discussed peculiari-

ties and difﬁculties of CS study constrain multiple re-

quirements on experts and developers, those have to

do not with the problem in question, but with theories

and formalisms from different disciplines.

Thus, the process of decision making and expert

systems design should be facilitated, and a portfolio

of methods and tools have to be offered to specialists

working with complex adaptive systems.

3 CONTEMPORARY

APPROACHES TO COMPLEX

SYSTEM STUDY

Recognition that a complex natural phenomenon has

to be studied using the principles of the system ap-

proach, induced some theoretical insights and their

practical realization. But, it should be said, that

many researchers still tend to use mono-discipline ap-

proaches and do not zoom in the complex emergent

behavior of CS and their causal-effect relations with

the environment.

Having revised a number of research works and

publications, we can mark out some general tenden-

cies for complex systems study.

Some research group offer ”island solutions”,

which, as a rule, are oriented to evaluation or as-

sessment of a few parameters or indicators, in other

words, they are dedicated to resolve speciﬁc goals.

For example, a system which provides indicator as-

sessment for a particular case study or for a limited

area. Domain ontologies for such systems are lim-

ited though may suffer from possible heterogeneity.

As a rule, such systems are effective when working

within the application domain and very sensitive to

any unforeseen changes (Karaca et al., 2009),(Urbani

and Delhom, 2005).

Multi-functional systems provide multiple analy-

sis of input information, can be based upon hybrid

techniques, possess tools and methods of data pre-

and post-processing, modelling, simulation. They are

less sensitive to changes in application domain, as

AGENT-BASED INTERDISCIPLINARY FRAMEWORK FOR DECISION MAKING IN COMPLEX SYSTEMS

they have tools to deal with uncertainty and hetero-

geneity (Nastar and Wallman, 2009), (Terry Bosso-

maier and Thompson, 2005).

Methodologies for software applications develop-

ment may support the mayor part of the system life

cycle phases, starting with the initial system planning,

which include system analysis and domain (prob-

lem) analysis phases, and then assist and provide EIS

design, coding, testing, implementation, deployment

and maintenance. In this case, consolidate cooper-

ation of specialists from various domains and with

various backgrounds is necessary (Gorodetski et al.,

2004), (Rotmans, 2006).

Basically, in spite of the diversity of existing ap-

proaches of DSS creation, obtained and reviewed out-

comes, it is not possible to elaborate a uniform overall

tool, capable of dealing with various domains and cre-

ate adequate solutions. However, the quality of solu-

tions and multi-function tools upsurges, because they

perform better results when these tools are oriented to

limited and speciﬁc domains.

Even then, exists a necessity to elaborate a general

methodology for DSS design oriented to distributed

heterogeneous domains, and powerful enough to fa-

cilitate work not only for small, but also for dis-

tributed and numerous research groups.

4 OUR APPROACH TO THE

FRAMEWORK CREATION

4.1 Interdisciplinary Approach

In general terms, a framework facilitates development

of an informational system, as it offers a consequence

of goals and works to do. In our case the informa-

tional system in the concept deﬁnition given above

(see part 2), represents a CS or a natural phenomenon.

Following the stages, determined in a framework,

developers and programmers will have more possibil-

ities to specialize in the domain of interest, meet soft-

ware requirements for the problem in term, thereby

reducing overall development time.

An effective framework should, on the one hand,

be general enough to be applicable for various do-

mains and, on the other hand, be adaptable for spe-

ciﬁc situations and problem areas. Moreover, the

framework should be based of the interdisciplinary

approach, which results from the melding of two or

more disciplines. Being applied to a complex sys-

tem, the approach includes the principal works: (1)

the complex system is decomposed into components

if necessary, the process of decomposition is repeated;

(2) then, each of the subsystems is studied by means

of the ”proper” techniques, belonging to the respected

discipline, or/and of hybrid methods; (3) managerial

process of decision making is realized, with feedback

and possible solutions generation.

As we have accentuated in the previous part, the

most important principles of CS is that they can not

be studies from a mono-discipline viewpoint, and it is

necessary to provide a complex hybrid application of

methods and techniques from various disciplines. Us-

ing intelligent agents seems to be an optimal solution

in this case (Weiss, 1999).

Actually, MAS helps to create cross-disciplinary

approaches for data processing, and, hence, for CS

study. An agent may include nontraditional instru-

ments to bear from different domains. Roles played

by an agent depend on the system (or subsystem)

functions and aims. There are no restrictions or limi-

tations put on knowledge and rule base, used by each

agent (Sokolova and Fern

andez-Caballero, 2009).

4.2 The Main Phases of The Framework

The purpose of our framework is to provide and fa-

cilitate complex systems analysis, simulation, and,

hence, their understanding and managing. From this

standpoint, and taking into account results and in-

sights, given in the previous parts, we implement the

principles of system approach.

The overall approach we use is straightforward:

we decompose the system into subsystems, and ap-

ply intelligent agents to study them, then we pool to-

gether obtained fragments of knowledge, and model

general patterns of the system behavioral tenden-

cies (Sokolova and Fern

andez-Caballero, 2007). The

framework consists of the three principal phases:

Preliminary Domain and System Analysis. This

is the initial and preparation phase when an analyst in

collaboration with experts study the domain of inter-

est, extract entities and discover their properties and

relations. Then, they state main and additional goal of

the research, possible scenarios and functions of the

system. During this exploration analysis, they search

answer to the questions: what the system has to do

and how it has to do it. As a result of this collabora-

tion it appears meta-ontology and knowledge base.

This phase is supported by the Protege Knowl-

edge editor, that implements the meta-ontology, and

by the Prometheus Design Kit, which we use to de-

sign multi-agent system and generate skeleton code

for further implementation of the DSS.

System Design and Codiﬁcation. The active ”el-

ement” of this phase is a developer. As supporting

ICAART 2010 - 2nd International Conference on Agents and Artificial Intelligence

software at this phase, we used the Jack

Intelligent

Agents and Jack

Development Environment. Once

the codiﬁcation is ﬁnished and the system is tested, th

second phase is completed.

Simulation and Decision Making. This is the last

phase of the framework and it has a special mission.

During this phase the ﬁnal users - decision makers -

can interact with the system, constructing solutions

and policies, estimating consequences of possible ac-

tions on the basis of simulation models.

5 THE CASE STUDY:

ENVIRONMENTAL IMPACT

ASSESSMENT ON HUMAN

HEALTH

5.1 Intelligent Agents

The general system organization is coherent to the

principal phases,described above.

Data search, retrieval, fusion and pre-processing

are realized by two intelligent agents: the Data Ag-

gregation agent (DAA) and the Data Preprocessing

agent (DPA), which do a number of tasks, following

the workﬂow:

Information Search - Data Source Classiﬁcation -

Data Fusion - Data Preprocessing - Believes Creation

Data mining procedures are based on the Func-

tion Approximation agent (FAA) and its team of

agents. The principal tasks to be solved here are: ”

to state the environmental pollutants that impact on

every age and gender group and determine if they are

associated with examined diseases groups; ” to cre-

ate the models which explain dependencies between

diseases, pollutants and groups of pollutants.

Here we are aimed to discover the knowledge in

form of models, dependencies and associations from

the pre-processed information which comes from the

previous logical layer. The workﬂow of this level in-

cludes the following tasks:

State Input and Output Information Flows - Cre-

ate models - Assess Impact - Evaluate models - Select

Models - Display The Results

Decision generation, simulation and human-

computer interaction are realized by the Computer

Simulation agent (CSA). The system works in a co-

operative mode, and it allows the decision maker to

modify, reﬁne or complete decision suggestions, pro-

vided by the system and validate them. This process

of decision improvement is repeated until the consol-

idated solution is generated. The workﬂow is repre-

sented below:

State Factors for Simulation - State the Values of

Factors - Simulate - Evaluate Results - Check Possible

Risk - Display The Results - Receive Decision Maker

Response - Simulate - Evaluate Results - Check Pos-

sible Risk - Display The Results

Agents communicate to each other and are trig-

gered by events and messages that they send. Agents

also share common data. Preliminary system design

was realized in the Prometheus Development Kit.

5.2 Methods used by Agents

The DAA has a number of subordinate agents un-

der its control; these are the Domain Ontology agent

and the fusion agents: the Water Data Fusion agent,

the Petroleum Data Fusion agent, the Mining Data

Fusion agent, the Trafﬁc Pollution Fusion agent, the

Waste Data Fusion agent and the Morbidity Data Fu-

sion agent. At the beginning of the execution the

DAA ﬁrstly send a message to the Domain Ontol-

ogy agent, which reads information from metaontol-

ogy, and returns in to the DAA. Then, the DAA starts

to searches for information sources and reviews them

trying to ﬁnd if there was a key ontological concept

there. If the ﬁle contains the concept, the Data Ag-

gregation agent sends an internal event to start data

retrieval to the specialized fusion agent. The plan

responsible for execution with the identiﬁed concept

starts reading the information ﬁle and searching for

terms of interest.

The DPA provides data preprocessing and has a

number of subordinate agents which specialize in dif-

ferent data clearing techniques: the Normalization

agent,the Correlation agent, the Data Smoothing

agent,the Gaps and Artifacts Check agent. They

perform all data preprocessing procedures, including

outliers and anomalies detection, dealing with miss-

ing values, smoothing, normalization, etc.

The FAA has a hierarchical team of subordinate

agents, which serve to support the roles: ”Impact As-

sessment”, ”Decomposition” and ”Function Approxi-

mation”. FAA has under its control a number of data

mining agents: the Regression agent, the ANN agent,

and the GMDH agent, which work in a concurrent

mode, reading input information and creating models.

Then, if any agent from this group ﬁnishes model-

ing, it calls for the Evaluation agent, which evaluates

obtained models, and return the list of the accepted

AGENT-BASED INTERDISCIPLINARY FRAMEWORK FOR DECISION MAKING IN COMPLEX SYSTEMS

ones, while the others are banned and deleted. The

FAA pools the output of the agents work, creates the

list with the accepted models and then, once the Re-

gression agent, the ANN agent, and the GMDH agent

ﬁnished their execution, calls for the Committee Ma-

chine agent, which creates the ﬁnal models in form of

committees for each of the dependent variables, and

validates them.

The methods which execute agents of the FAA

team are the following:

• Statistical modelling - include linear, non-linear

simple and multiple models and their evaluation.

• Artiﬁcial Networks modelling - include feed-

forward neural networks, trained by back-

propagation algorithm, genetic algorithms, radial-

based function networks.

• Group Method of Data Handling (GMDH),

which is one of the powerful methods of self-

organization. We used the combinatorial algo-

rithm of GMDH (Madala, 1994).

• Committee Machines, which provide creation of

weighted hybrid models with the condition that

the best models selected for the committee have

more impact on the ﬁnal result (Haykin, 1998).

The CSA interacts with user and performs a set of

tasks within ”Computer Simulation”, ”Decision Mak-

ing” and ”Data Distribution” roles. It has the agent

team, which includes Forecasting agent, Alarm agent

and View agent.

The process of simulation and generation of pos-

sible solutions is interactive. The human-computer

interaction sessions are organized by the View agent,

which offer window -forms, graphical and textual

documentation as supporting tools. The Forecasting

agent calculated predicting values for the CSA, and

the Alarm agent checks if these values satisfy permit-

ted standards. Figure 1 gives a look at agent teams.

5.3 Simulation Results

The MAS has an open agent-based architecture,

which would allow us an easy incorporation of addi-

tional modules and tools, enlarging a number of func-

tions of the system. The system belongs to the orga-

nizational type, where every agent obtains a class of

tools and knows how and when to use them. Actually,

such types of systems have a planning agent, which

plans the orders of the agents’ executions. In our case,

the main module of the JACK

program carries out

these functions.

The View agent displays the outputs of the system

functionality and realize interaction with the system

Table 1: Part of the Table with the outputs of impact assess-

ment.

N Disease Pollutant

1 Neoplasm Nitrites

in water;

Miner products;

DBO5; Asphalts;

Dangerous

chemical wastes;

Fuel-oil; Petroleum

liquid gases;

Water: solids

in suspension;

Non-dangerous

chemical wastes;

2 Diseases of DBO5;

Miner products;

the blood and Fuel-oil;

blood- forming Nitrites in water;

organs, Solids in

the immune suspension;

mechanism Dangerous

metallic wastes

Miner products;

3 Pregnancy, Kerosene;

childbirth and Petroleum

the puerperium liquid gases;

Solids in suspension;

Petroleum; DQO;

Fuel-oil; Asphalts;

Gasohol; DBO5;

Water: Nitrites.

Petroleum;

Petroleum autos;

user. As the system is autonomous and all the calcu-

lations are executed by it, the user has only access to

the result outputs and the simulation window.

To evaluate the impact of environmental param-

eters upon human health in Castilla-La Mancha, in

general, and in the city of Albacete in particular, we

have collected retrospective data since year 1989, us-

ing open information resources offered by the Spanish

Institute of Statistics and by the Institute of Statistics

of Castilla-La Mancha.

As indicators of human health and the inﬂuencing

factors of the a environment, which can cause nega-

tive effect upon the noted above indicators of human

health, the levels of contamination of potable water,

air, soil, and indicators of trafﬁc activity, hazardous

wastes, energy and used annually, etc.

The DSS has recovered data from plain ﬁles,

which contained the information about the factors of

interest and pollutants, and fused in agreement with

the ontology of the problem area. It has supposed

some necessary changes of data properties (scalabil-

ity, etc.) and their pre-processing. After these proce-

dures, the number of pollutants valid for further pro-

ICAART 2010 - 2nd International Conference on Agents and Artificial Intelligence

100

Figure 1: The agent teams created for the case study.

cessing has decreased from 65 to 52. This signiﬁcant

change was caused by many blanks related to several

time series, as some factors have started to be regis-

tered recently.

After considering this as an important drawback,

it was not possible to include them into the analysis.

The human health indicators, being more homoge-

neous, have been fused and cleared successfully. The

impact assessment has shown the dependencies be-

tween water characteristics and neoplasm, complica-

tions of pregnancy, childbirth and congenital malfor-

mations, deformations and chromosomal abnormali-

ties.

The MAS has a wide range of methods and tools

AGENT-BASED INTERDISCIPLINARY FRAMEWORK FOR DECISION MAKING IN COMPLEX SYSTEMS

101

Figure 2: (a) The ﬁnal model for one of the diseases of the case study. (b) The outcomes of the impact assessment for one of

the diseases of the case study.

for modeling, including regression, neural networks,

GMDH, and hybrid models, based on committee ma-

chines. The function approximation agent selected

the best models, which were: simple regression -

4381 models; multiple regression - 24 models; neu-

ral networks - 1329 models; GMDH - 2435 models.

The selected models were included into the commit-

tee machines. We have foretasted diseases and pollu-

tants values for the period of four years, with a six

month step, and visualized their tendencies, which

are going to overcome the critical levels. Control un-

der the ”signiﬁcant” factors, which cause impact upon

health indicators, could lead to decrease of some types

of diseases.

Committee machines provide universal approxi-

mation, as the responses of several predictors (ex-

perts) are combined by means of a mechanism that

does not involve the input signal, and the ensemble

average value is received. As predictors we used re-

gression and neural network based models.

An example of the ﬁnal model, received by com-

mittee machine for the ”‘Neoplasms”’ for the region

of Castilla-La Mancha, Spain, is given on Figure 2a,

and the results of the impact assessment are shown on

Figure 2b.

6 CONCLUSIONS

Complex system analysis and modelling is still a com-

plicated problem, and an efﬁcient framework which

supports decision making and expert systems creation

facilitates research efforts for various research groups.

The framework we have demonstrated in the article, is

one of the approximations to solve this complicated

task. And, to conclude with, we should note some

essential advantages we have reached, and some di-

rections for future research.

First, the framework is interdisciplinary and is

ﬂexible for changes: it can be adapted to comply with

speciﬁc features of the domain of interest. The pro-

totype of the DSS, created in accordance with the

framework, supports decision makers in choosing the

behaviour line (set of actions) in such a general case,

which is potentially difﬁcult to analyze and foresee.

As for any complex system,the human choice is deci-

sive.

Second, in spite of our time consuming modeling

work, we are looking forward to both revising and im-

proving the system and deepening our research.

Third, we consider making more experiments

varying as with data structure, trying to apply the sys-

tem to the other.

ICAART 2010 - 2nd International Conference on Agents and Artificial Intelligence

102

The framework supports all the necessary steps for

standard decision making procedure by utilizing intel-

ligent agents. The teams of intelligent agents, that are

logically and functionally connected, have been pre-

sented. Real-time interaction with the user provides

a range of possibilities in choosing one course of ac-

tion from among several alternatives, which are gen-

erated by the system through guided data mining and

computer simulation. The system is aimed to regular

usage for adequate and effective management by re-

sponsible municipal and state government authorities.

We used both traditional data mining techniques,

hybrid and speciﬁc methods, with respect to data na-

ture (incomplete data, short data sets, etc.). Combina-

tion of different tools let us gain in quality and preci-

sion of the reached models, and, hence, in recommen-

dations, which are based on these models. Received

dependencies of interconnections and associations be-

tween the factors and dependent variables help to cor-

rect recommendations and avoid errors.

ACKNOWLEDGEMENTS

This work was partially supported by Spanish Min-

isterio de Ciencia e Innovacion TIN2007-67586-C02-

02, and Junta de Comunidades de Castilla-La Mancha

PII2I09-0069-0994 and PEII09-0054-9581 grants.

REFERENCES

Athanasiadis, I. N. and Mitkas, P. A. (2004). An agent-

based intelligent environmental monitoring system.

CoRR, cs.MA/0407024.

Gorodetski, V. I., Karsaev, O., Samoilov, V., Konushy, V.,

Mankov, E., and Malyshev, A. (2004). Multi-agent

system development kit. In Intelligent Information

Processing.

Haykin, S. (1998). Neural Networks: A Comprehensive

Foundation. Macmillan, New York.

Karaca, F., Anil, I., Alagha, O., and Camci, F. (2009).

Trafﬁc related pm predictor for besiktas, turkey. In

Athanasiadis, I. N., Mitkas, P. A., Rizzoli, A. E.,

and G

omez, J. M., editors, ITEE, pages 317–330.

Springer.

Levin, M. S. (2006). Composite Systems Decisions (Deci-

sion Engineering). Springer-Verlag New York, Inc.,

Secaucus, NJ, USA.

Madala, H.R., I. A., editor (1994). Inductive Learning Al-

gorithms for Complex Systems Modelling. CRC Press

Inc., Boca Raton, Ann Arbor, London, Tokyo.

Nastar, M. and Wallman, P. (2009). An interdisciplinary

approach to resolving conﬂict in the water domain. In

Information Technologies in Environmental Engineer-

ing Proceedings of the 4th International ICSC Sympo-

sium Thessaloniki, Greece.

Rechtin, E. (1999). Systems architecting of organizations:

why eagles can’t swim. CRC Press.

Ria

no, D., S

anchez-Marr

e, M., and R.-Roda, I. (2001). Au-

tonomous agents architecture to supervise and control

a wastewater treatment plant. In IEA/AIE.

Rotmans, J. (2006). Tools for integrated sustainability as-

sessment: A two-track approach. Integrated Assess-

ment, 6(4).

Sokolova, M. V. and Fern

andez-Caballero, A. (2007). A

multi-agent architecture for environmental impact as-

sessment: Information fusion, data mining and deci-

sion making. In ICEIS 2007 - Proceedings of the

Ninth International Conference on Enterprise Infor-

mation Systems, Volume AIDSS, Funchal, Portugal.

Sokolova, M. V. and Fern

andez-Caballero, A. (2008). Fa-

cilitating mas complete life cycle through the prot

prometheus approach. In Agent and Multi-Agent Sys-

tems: Technologies and Applications, KES-AMSTA,

Incheon,Korea.

Sokolova, M. V. and Fern

andez-Caballero, A. (2009). Data

mining driven decision making. In ICAART 2009

- Proceedings of the International Conference on

Agents and Artiﬁcial Intelligence, Porto, Portugal.

Sokolova, M. V. and Fern

andez-Caballero, A. (2009). Mod-

eling and implementing an agent-based environmental

health impact decision support system. Expert Syst.

Appl., 36(2):2603–2614.

Terry Bossomaier, Denise Jarratt, M. M. A. T. S. and

Thompson, J. (2005). Data integration in agent based

modelling. Complexity International, 11.

Urbani, D. and Delhom, M. (2005). Water management

policy selection using a decision support system based

on a multi-agent system. In AI*IA.

Weiss, G., editor (1999). Multiagent systems: a modern

approach to distributed artiﬁcial intelligence. MIT

Press, Cambridge, MA, USA.

AGENT-BASED INTERDISCIPLINARY FRAMEWORK FOR DECISION MAKING IN COMPLEX SYSTEMS

103