A MULTI-AGENT ARCHITECTURE FOR ENVIRONMENTAL

IMPACT ASSESSMENT

Information Fusion, Data Mining and Decision Making

Marina V. Sokolova

1,2

and Antonio Fernández-Caballero

Universidad de Castilla-La Mancha, Departamento de Sistemas Informáticos &

Instituto de Investigación en Informática de Albacete, Campus Universitario s/n, 02071-Albacete, Spain

Kursk State Technical University, ul.50 let Oktiabrya, 94, Kursk, 305040, Russia

Keywords: Environmental impact, Multi-agent system, Data fusion, Data handling, Conceptual hierarchy, Decision

support system.

Abstract: The paper introduces an approach to creating a multi-agent architecture for environmental impact

assessment upon human health. As the indicators of the environmental impact we assume water pollution,

indexes of traffic and industrial activity, wastes and solar radiation; and as the human health indicator we

take morbidity. All the data comprise multiple heterogeneous data repositories. The general structure of the

architecture is represented. Thus, the proposed system is logically and functionally divided into three layers,

solving the tasks of information fusion, pattern discovery through data mining, and decision support

making, respectively, which are discussed. The discovered patterns will be used as a foundation for real-

time decision making, which should be of great importance for adequate and effective management by

responsible municipal and state government authorities.

1 INTRODUCTION

The convenience of our research is stated by the fact

that environmental pollution as a result of energy

production, transportation, industry, or lifestyle

choices adversely affects health. The term

“environmental pollution” includes such factors as

ambient and indoor air pollution, water pollution,

inadequate waste management, pesticides, noise and

radiation (Turunen & Latola, 2005; Carrillo &

González-Chávez, 2006). In addition, people usually

face a deteriorated environment which affects their

health and provokes its degradation within a

population due to their life styles and the aggressive

ecological impact. This is demonstrated by

increasing number of endogenous diseases (such as

birth defects, chromosome diseases, etc) and some

classes of exogenous diseases (diseases of the skin

and subcutaneous tissue, endocrine and metabolic

diseases, neoplasm and some others).

The complete set of pollution and health data

forms a complex system, which inhabits all the

necessary characteristics to de modeled by means of

multi-agent systems (MAS) approach, e.g.

modularity, decentralization, changeability, ill

structure and weak predictability (Bradshaw, 1997;

Wooldridge, 2002; López-Jaquero et al., 2005).

Multi-agent approach is to our opinion the best

technique that can help reducing the complexity of

the system by creating module components, which

solve private subtasks, constituting together the

whole goal.

In this paper we introduce our proposal for a

multi-agent architecture for environmental impact

assessment, structured into three levels: Information

Fusion, Data Mining and Decision Making.

2 RELATED WORKS

Multi-agent systems have been in the center of

active research for more than ten years and resulted

in many successful applications. There is a range of

works dedicated to environment and human health,

as described next. Furthermore, the application of

Data Mining (DM) techniques for environmental

monitoring, medicine, social issues is also quite

common. In one of the related works (Athanasiadis

& Mitkas, 2004) it is reported about applying the

219

V. Sokolova M. and Fernández-Caballero A. (2007).

A MULTI-AGENT ARCHITECTURE FOR ENVIRONMENTAL IMPACT ASSESSMENT - Information Fusion, Data Mining and Decision Making.

In Proceedings of the Ninth International Conference on Enterprise Information Systems - AIDSS, pages 219-224

DOI: 10.5220/0002346802190224

 SciTePress

software agent paradigm to environmental

monitoring informational system embodied by

MAS. In another paper (Gorodetski et al., 2005) an

approach to agent-based situation assessment system

development for security-related applications is

presented. DM techniques for knowledge

discovering and early diagnostics were utilized to

early intervention to developmentally-delayed

children (Chang, 2006). In the work by Chen and

Bell (2002) the MAS, which is aimed to reveal

correlations between human health and

environmental stress factors (traffic activity,

meteorological data and noise monitoring

information) using wide range of DM methods is

reported about.

Although all the works have demonstrated novel

and promising practical and theoretical outcomes, it

seems to be important to create a MAS for

knowledge discovering and assessment of

environmental tension upon the population by detail

analysis of endogenous and exogenous diseases

cases.

3 THE PROBLEM AREA

The main practical aim of the project is to create an

agent-based system for state situation assessment,

monitoring the environment pollution and following

the corresponding changes in human health,

generating a set of alternatives for successful and

sustainable situation management.

Continuous processing and maintenance of the

information requires essential efforts from the

practitioners and professionals not only while

handling and storing data, but fundamentally when

interpreting it. Actually, it seems very hard to handle

all the data without using DM methods, which can

autonomously dig out all the valuable knowledge

that is embedded in a database without human

supervision, providing a full life-cycle support of

data analysis.

Working with public health information puts on

restrictions caused by the methodologies of data

measurement, the standards currently in use, data

availability, etc. For instance, it is known that

International Statistical Classification of Diseases

and Related Health Problems (ICD) was reviewed

10 times, International Classification of Functioning

and Disability (ICIDH) – 2 times, and local

standards were also reviewed relatively.

In recent years, the tendency to use products and

energy life-cycle indicators in order to assess

ecological impact has appeared. This approach

seems to be effective when evaluating quota of

industrial, chemical and traffic activity impact and

we accept to follow it in our work.

The concept hierarchy was created using

ontology editor Protégé 3.2, and includes the

information about the regions of interest and the

examining indicators of our current study. The

ontology contains the diseases classes in accordance

with the ICD-10 and environmental pollution

indicators: water pollution, dangerous wastes,

transport activity, and industrial activity parameters

revealing dangerous emissions during energy life-

cycle. All this has been detailed by years and other

sub-indexes. In the ontology we have made accent in

regions, which are characterized with some

environmental pollution and human health level.

4 SYSTEM ARCHITECTURE

Fig.1 shows the layout of the model where system

goals and agent roles are indicated. The system

consists of three levels; the first is aimed for meta-

data creation, the second one is responsible for

hidden knowledge discovering, and the third level

provides real-time decision support making, data

distribution and visualization.

4.1 Analysis of the System with Gaia

Methodology

There are many alternative agent-oriented software

engineering methodologies, including MaSE

(DeLoach et al., 2001), Gaia (Wooldridge, 2000),

Agent ULM, Prometheus (Padgham and Winikoff,

2002), Tropos (Giunchiglia et al., 2002), INGENIAS

(Gómez-Sanz & Pavón, 2003) and some others. In

our study analysis was performed following Gaia

methodology.

The analysis has led to the identification of two

roles on the first level, three roles on the second

level and also three roles on the third level of the

MAS. The roles show the detailed functionality of

the system. Agents are responsible for execution of

the extracted activities.

All the protocols are named in a similar way to

indicate that they carry out related functions

consisting in transmission of processed data and

information from one agent to another. The safety

responsibilities for the roles are specified by means

of a list of predicates and states that the activities

have to bring up as results and notify them.

ICEIS 2007 - International Conference on Enterprise Information Systems

220

Support in

Decisiуn

Making

Knowledge

discovery

Meta-data

creation

Level 3– “Decision making”

Simulation, Decision making, Alarm generation and Data Distribution

Simulation agents, Decision making agents, Distribution agents

Level 1 – “Information Fusion”

Data Aggregation and Clearing

DB handling agents, DB aggregation agents

Level 2 – “Data Mining”

Impact assessment, Decomposition and Function Approximation

Analysing agents, Evaluation agents

user

web

document

Expert

knowledge

Data Sources

Decision

maker

preferences

Figure 1: The general system schema with main tasks and agent roles.

4.1.1 Level 1: Information Fusion

The first level, as detailed in Fig. 2, is named

“Information fusion” and it acquires data from

diverse sources and in different format types. The

input to this level is through protocol ReturnEI,

which incorporates all the expert knowledge. The

aims of this level include data monitoring,

validation, clearing and fusing into a common meta-

data scheme. These tasks are grouped into two roles:

“Data Fusion” and “Data Clearing”. There are two

general types of agents at this level: DB handling

agents and DB aggregation agents (see Fig.1, level

1), fully controlling data maintenance and executing

all the necessary pre-processing functions at the

every step.

There are two general types of agents at this

level: DB handling agents and DB aggregation

agents (see Fig.1, level 1), fully controlling data

maintenance and executing all the necessary pre-

processing functions at the every step.According to

the scheme of level 1, firstly we fuse incoming raw

data and form a meta-data base, consisting in time

series, which then pass throw the sequential data

processing steps: noise reduction, outlier elimination

and doubling, and inconsistent and missing values

checking. The role “Data Fusion” supposes the

following logical steps:

 Vocabulary creation: Create a vocabulary for

the domain of interest.

 Hierarchy creation: Create a conceptual

hierarchy by assigning weights to each class

of concepts and by determining relations

among classes.

 DB transformation with Ontology Algebra:

Select data from different sources with respect

to hierarchy using ontology algebra.

 Meta-data base aggregation: Combine data

together.

Then protocol ReturnDF delivers meta-data base

to agents of role “Data Clearing”, that is to say, to

DBHandling Agents (see Fig.1, level 1). The

DBHandling Agents check data for outliers, smooth

time series and interpolate missing values with

weighted moving averages.

The final meta-data base consists of sequences of

ordered indicator values, measured at equal time

intervals (time-series). It is delivered to the next

layer for knowledge discovering through the

ReturnDC protocol.

A MULTI-AGENT ARCHITECTURE FOR ENVIRONMENTAL IMPACT ASSESSMENT: Information Fusion, Data

Mining and Decision Making

221

Level 1

Role Schema

Data Clearing

Role Schema

Data Fusion

Vocabulary

creation

Hierarch

creation

DB transformation

with Ontology

Algebra

Protocol

ReturnEI

Expert

Informa

tion

expert

Protocol

ReturnDC

to Level 2

Meta-data base

aggregation

Protocol

ReturnDF

Outliers

clearing

Weighted

moving

avera

Missing

Values

correction

Figure 2: Layout of the first level.

Level 2

Role Schema

Decomposition

into groups

Correlation

Analysis

Protocol

ReturnD

Role Schema

Function Approximation

Models

acceptance

Library of Data Mining methods

Data Mining method 1

Data Mining method 2

Data Mining method N

. . . . .

Approximation

acceptance

Role Schema

Impact Assessment

PCA

to Level 3

Protocol

ReturnFA

Protocol

ReturnIA

Protocol

ReturnDC

Figure 3: Layout of the second level.

4.1.2 Level 2: Data Mining

The second level is responsible for knowledge

mining from meta-data base (Fig. 3.). There will be

a number of techniques to can be used for data

analysis, all of them provided by Analysis agents.

There are three roles at this level: “Decomposition”,

“Impact Assessment” and “Function

Approximation”, which will be performed by

Analyzing agents and Evaluation agents (as

previously shown in Fig.1, level 2).

Through the role “Impact Assessment”, which

apply the procedure of principal component analysis

(PCA), we aim to reveal interconnections between

health and pollution indicators and qualitatively

evaluate the influence of the latter. The procedure is

sequentially delivered for every class of diseases and

to the totality of environmental pollution indicators

(Sokolova, Rashad & Skopin, 2006).

The other roles solved at level 2 are

“Decomposition” and “Function approximation”.

We will require models for computer simulation,

forecasting and decision making. In order to

decrease the number of simultaneously processed

indicators and be saved from intercorrelation and

multi-colinearity between them, Analyzing agents

initiate a procedure of factor space decomposition by

calculating the correlation matrix and its further

decomposition (Artemenko, et al, 2004). We will

receive a set of independent variables and those that

do not correlate significantly between them and with

certain class of diseases and can be used as factors

for modeling this class of diseases. The protocol

ReturnD transfers information about decomposition

to Analyzing agents.

Then, for every indicator we will extract models

revealing their tendencies by Analyzing agents,

executing the procedure of function approximation,

are based on different methods, which are stored in

ICEIS 2007 - International Conference on Enterprise Information Systems

222

the library of DM methods. For example, for the role

“Function Approximation” several kinds of agents

may be called: statistical ones utilizing regression

modeling for activities DataMiningMethod1 and

DataMiningMethod2, and decision trees techniques

for activity DataMiningMethod3, and another

technique for activity DataMiningMethodN. The

agents execute in interleaved mode. As a result, we

have several different models for every indicator and

we will choose the best one.

The Evaluation agents check if the models are

adequate to data sets. Then Evaluation agents select

the models which best fulfill the requirements. All

the results of data transformations are distributed to

the next level for decision making through protocols

ReturnIA and ReturnFA.

4.1.3 Level 3: Decision Making

The third level carries out a set of procedures

including model evaluation, computer simulation,

decision making and forecasting based on the

models created on the previous level (ReturnFA) and

the results of the PCA (ReturnIA).

The structural schema of this level is given in

Fig. 3. There are three roles: “Computer Simulation

and Forecasting”, “Decision Making” and “Data

Distribution”. Protocols ReturnFA (which delivers

the output of “Function Approximation”) and

ReturnIA (delivering results of “Impact

Assessment”) come from the previous level.

Protocols ReturnCS and ReturnDM deliver

simulation and decision making results for further

processing. The end-user (person making decisions)

interacts with the MAS through the SUI (System-

User Interaction) protocol. The user chooses the

indicator he wants to examine and initiates a

computer simulation.

He may choose the independent variables and

state their values and forecasting period. Then the

information is delivered to Simulation agents (see

Fig.1, level 3) that perform computations for every

model and repeats it until the outputs for all the

models are received. Then SUI protocol delivers

results to the end-user for future decision making.

The Decision making agents (see Fig.1, level 3),

in agreement with their believes, recommend the

most optimal variants of computer simulation, which

are selected by standard decision making criteria

(Bayes, minimax, Hurvitz, etc) to the user. Decision

making agents also control the forecasted values. In

case they exceed or are likely to exceed the

permissible levels, an alarm message will be

generated, visualized and sent to the user.

At the last step, the information is delivered to

the final destination – end-users and applications in

the form of web messages, textual files, e-mails and

visual presentations. This level transforms the

revealed information including results of computer

simulation, forecasting and decision making into

understandable and multiple forms. These tasks are

realized by the Distribution agents, which operate

on combining textual and graphical descriptions of

recommendations.

User

Level 3

Protocol

SUI

Protocol

ReturnIA

Protocol

ReturnFA

Role Schema

Decision Making

Protocol

ReturnCS

Role Schema

Computer Simulation

and Forecasting

Protocol

ReturnDM

Role Schema

Data Distribution

Generate alarm alert

Making forecast

Simulating every

model

Stating values for

simulation

Choising the models

Calculating criterias

Create module with

textual and graphical

information

Create document

Create web-

Create e-mail

Figure 4: Layout of the third level.

A MULTI-AGENT ARCHITECTURE FOR ENVIRONMENTAL IMPACT ASSESSMENT: Information Fusion, Data

Mining and Decision Making

223

5 CONCLUSIONS AND FUTURE

WORK

In this paper we have introduced an approach to

developing an intelligent environmental situation

monitoring and evaluation decision support through

MAS, which uses software and works with

heterogeneous data sources. We discussed the nature

and peculiarities of experimental data and expert

knowledge used in our system, described an

ontology and presented a general system

architecture. In accordance with requirements of

Gaia methodology we extracted and explained in

detail the roles and associated set of interactions.

The supposed approach to environmental impact

assessment through multy-agent system enables to

identify and evaluate quantitatively which certain

type of pollutants affects health, approximate and

forecast the tendencies of situation development and

allows a user to exploit the inherent potentialities of

real-time simulation. The software agents use data

mining methods for knowledge discovery, which

will be used as a foundation for support in decision

making and recommendation generating. This

should be of great importance for adequate and

effective management by responsible municipal and

state government authorities.

The system developed is being used as a pilot

project in Spanish University of Castilla-La Mancha

and Institute of Regional Development of Albacete.

In our future work we will concentrate on working

out the MAS and its implementation into practical

use.

ACKNOWLEDGEMENTS

Marina V. Sokolova is the recipient of a

Postdoctoral Scholarship (Becas MAE) awarded by

the Agencia Española de Cooperación Internacional

of the Spanish Ministerio de Asuntos Exteriores y de

Cooperación.

REFERENCES

Agent UML: http://www.auml.org/

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

intelligent environmental monitoring system. In:

Management of Environmental Quality, 15, 238-249.

Artemenko, M., Hudec, O., Lapina, T. and Sokolova,

M.V, (2004). The information-analytical program tool

for regional decision making in social sphere. In:

Telecommunications, 9, 42-44.

Bradshaw, J.M. (1997). Software Agents. The MIT Press.

Carrillo González, R., and González-Chávez, M.C.A.

(2006). Metal accumulation in wild plants surrounding

mining wastes. Environmental Pollution, 144, 84-92.

Chang, C.L. (2006). A study of applying data mining to

early intervention for developmentally-delayed

children. Expert Systems with Applications. In press.

Chen, H., Bell, M. (2002). Instrumented city database

analysts using multi-agents. Transportation Research,

Part C, 10, 419–432.

DeLoach, S.A., Wood, M.F., and Sparkman, C.H. (2001).

Multiagent systems engineering. International Journal

of Software Engineering and Knowledge Engineering,

11, 231-258

Giunchiglia, F., Mylopoulos, J., and Perini, A. (2002). The

Tropos software development methodology:

Processes, models and diagrams. In: Third

International Workshop on Agent-Oriented Software

Engineering, Juna.

Gómez-Sanz, J., and Pavon, J. (2003). Agent oriented

software engineering with INGENIAS. Lecture Notes

in Computer Science, 2691 394–403.

Gorodetsky, V., Karsaeyv, O., Samoilov, V. (2005).

Multi-agent and data mining technologies for situation

assessment in security-related applications. In: Dunin-

Keplicz, B., Jankovski, A., Skowron, A., and Szczuka,

M. (eds.), Monitoring, Security, and Rescue

Techniques in Multi-agent Systems, 411-422.

International Classification of Diseases (ICD):

http://www.who.int/classifications/icd/en/

López-Jaquero, V., Montero, F., González, P., and

Fernández-Caballero, A. (2005). A multi-agent system

architecture for the adaptation of user interfaces.

Lecture Notes in Artificial Intelligence, 3690, 583-586.

Padgham, L. and Winikoff, M. (2002). Prometheus: A

pragmatic methodology for engineering intelligent

agents. In: Proceedings of the OOPSLA 2002

Workshop on Agent-Oriented Methodologies, Seattle.

Protégé: http://protege.stanford.edu/

Sokolova, M.V., Rashad J. Rasras, and Skopin, D. (2006).

The artificial neural network based approach for

mortality structure analysis. In: American Journal of

Applied Science, 3, 1698-1702.

Turunen, M., and Latola, K. (2005). UV-B radiation and

acclimation in timberline plants. Environmental

Pollution, 137, 390-403.

Wooldridge, M. (2002). An Introduction to Multiagent

Systems. John Wiley & Sons.

Wooldridge, M., Jennings, N.R., Kinny, D. (2000). The

Gaia Methodology for Agent-Oriented Analysis and

Design. Journal of Autonomous Agents and Multi-

Agent Systems, 3, 285-312.

ICEIS 2007 - International Conference on Enterprise Information Systems

224