method dex: a qualitative, hierarchical, rule-based...

Method Method DEX:DEX:

A QualitativeA Qualitative, , HierarchicalHierarchical, , RuleRule--Based Approach Based Approach

to to MultiMulti--Criteria Decision ModellingCriteria Decision Modelling

Method Method DEX:DEX:

A QualitativeA Qualitative, , HierarchicalHierarchical, , RuleRule--Based Approach Based Approach

to to MultiMulti--Criteria Decision ModellingCriteria Decision Modelling

Marko BohanecMarko Bohanec

Jožef Stefan Institute, Department of Knowledge Technologies, Ljubljana, Slovenia

and

University of Nova Gorica, Nova Gorica, Slovenia

• JSI is a leading national research organization in natural sciences and technology

• Founded in 1949

• Named after a distinguished physicist Jožef Stefan (1835-1893)

• Staff: about 700 researchers

• Research areas:

• physics

• chemistry, biochemistry & nanotechnology

• nuclear technology

• information and communication sciences & electronics

Jožef Stefan InstituteJožef Stefan Institute

DepartmentDepartment ofof KnowledgeKnowledge TechnologiesTechnologies

Head: Prof. Nada Lavrač

Staff: about 45

Research Areas:

Decision SupportDecision Support

• Methods – Decision Analysis

– Decision Modelling

• Qualitative Multi-Attribute Modelling

• Combined with Data Mining

– Decision Support Systems

• Software – DEXi: Multi-attribute decision support

• Applications: – Applications for clients: evaluation of

companies, personnel management, project evaluation, land-use planning, agronomy, medicine and health-care, sports, banking and finance

Hormonalcircumstances

Personalcharacteristics Other

Menstrualcycle Fertility

Oralcontracept.

RISK

Cancerog.exposure

Fertilityduration

Reg. andstab. of men.

Age

First delivery

# deliveries

Quetel'sindex

Familyhistory

Demograph.circumstance

Physicalfactors

ChemicalfactorsMenopause

OutlineOutline

DEX (Decision EXpert):

• Motivation, purpose and historical context

• Methodology:

– Static aspects: decision model components, representation

– Dynamic aspects: model creation, modification, utilization

• Applications and software

• Research problems and future work

What is DEX?What is DEX?

DEX

Multi-Criteria Decision Analysis

• modeling using criteria and

utility functions

• problem decomposition and

structuring

• evaluation and analysis of

decision alternatives

Multi-Criteria Decision Analysis

• modeling using criteria and

utility functions

• problem decomposition and

structuring

• evaluation and analysis of

decision alternatives

Artificial Intelligence

Expert Systems

• qualitative (symbolic) variables

• "if-then" rules

• decision model = knowledge base

• handling imprecision and uncertainty

• transparent models, explanation

Machine Learning

Artificial Intelligence

Expert Systems

• qualitative (symbolic) variables

• "if-then" rules

• decision model = knowledge base

• handling imprecision and uncertainty

• transparent models, explanation

Machine Learning

Fuzzy sets

• verbal measures

• fuzzy operators

Fuzzy sets

• verbal measures

• fuzzy operators

What is DEX?What is DEX?

Method characteristics:

1. Multi-Attribute (Multi-Criteria):

Evaluates alternatives through aggregation of multiple criteria

2. Hierarchical:

Attributes are structured hierarchically (as in AHP or MCHP)

3. Qualitative:

Attributes are discrete, verbal (e.g. “low”, “med”, “high”)

4. Rule-based:

Aggregation is defined by decision rules in decision tables

DEXDEX

Method for qualitative multi-attribute modeling

DEX is similar to other “full aggregation” multi-attribute methods:

1. Multiple attributes, hierarchically structured

2. Evaluation of alternatives: bottom-up aggregation

CAR CAR

TECH.CH. TECH.CH. PRICE PRICE

COMFORT COMFORT SAFETY SAFETY MAINT MAINT BUYING BUYING FUEL FUEL

Some Car Some Car

DEXDEX


DEX is different from the majority of multi-attribute methods:

1. Attributes are discrete, symbolic, qualitative

CAR CAR



BUYING {high, medium, low}

FUEL {low, medium, high}

TECH.CH. {bad, acc, good, exc}

DEXDEX


DEX is different from the majority of multi-attribute methods:

1. Attributes are discrete, symbolic, qualitative

Attribute scales can be unordered (categorical),

but are typically preferentially ordered (increasing or decreasing) “criteria”

CAR CAR



BUYING {high, medium, low}

FUEL {low, medium, high}

TECH.CH. {bad, acc, good, exc}

DEXDEX


DEX is different from other multi-attribute methods:

2. Evaluation of alternatives (aggregation) is defined by decision tables

CAR CAR


COMFORT COMFORT SAFETY SAFETY

MAINT MAINT BUYING BUYING

FUEL FUEL

FUEL SAFETY COMFORT TECH.CH.

high good exc unacc

low bad med unacc

... ... ... ...

med good med good Elementary decision rule:

if FUEL=med & SAFETY=good and COMFORT=med

then TECH.CH.=good

DEX Method: HistoryDEX Method: History

1980 2000 1990 2010

DECMAK

Methodology

• further improvement

Software

• DEXi

Education

International applications

• Sol-Eu-Net

• agriculture, food, GMO

• project evaluation

• finance

• medicine & health care

Related

• model revision, proDEX

DEX DEXi

Methodology

• initial development

Software

• DECMAK

• “toolbag”

First applications

• HW and SW

selection

• personnel mgmt

• nursery schools

Methodology

• integration

Software

• DEX

• Vredana

National applications

• Housing Fund

• Ministry Sci-Tech

• Talent System

• industry

• medicine

Related

• HINT

OutlineOutline



• Methodology:





DEX Model: Formal RepresentationDEX Model: Formal Representation


basic attributes - inputs

aggregate attributes


basic attributes - inputs

aggregate attributes

outputs


Possible formulations of output range Ei:

“Basic” “Practical” “Extended”

Ei ≡ Di Ei ≡ interval over Di

Ei ≡ distribution over Di

(probabilistic, fuzzy)

acc 0.2

good 0.7

exc 0.1

DEX Method: Dynamic AspectsDEX Method: Dynamic Aspects

How to:

• Obtain model and its

components?

• Modify, edit, maintain

the model?

• Verify model and its

components (e.g. for

completeness and

consistency)?

• Deal with uncertainty?

• Ensure transparency,

comprehensibility?

How to:

• Obtain model and its

components?

• Modify, edit, maintain

the model?

• Verify model and its

components (e.g. for

completeness and

consistency)?

• Deal with uncertainty?

• Ensure transparency,

comprehensibility?

Creation Usage

How to:

• Obtain and represent

data about alternatives?

• Deal with incomplete,

uncertain data?

• Evaluate alternatives?

• Analyze the results?

• Explain and justify

results?

• Validate results?

• Assess the quality of

decision?

How to:

• Obtain and represent

data about alternatives?

• Deal with incomplete,

uncertain data?

• Evaluate alternatives?

• Analyze the results?

• Explain and justify

results?

• Validate results?

• Assess the quality of

decision?

DEXiDEXi:: Program for MultiProgram for Multi--Attribute Decision MakingAttribute Decision Making

Functionality

• creation and editing of qualitative DEX models:

– model structure

– decision tables

• acquisition and evaluation of alternatives

• analysis of alternatives: “what-if”, “±1 analysis”, comparison of alternatives, selective explanation

• tabular and graphical reports

http://kt.ijs.si/MarkoBohanec/dexi.html



Obtaining attributes, their value scales and model structure:

– Expert modeling, ‘hand-crafting’,

following guidelines and ‘rules of thumb’

– Machine learning from data (methods: HINT, Model Revision)

Creation


Acquisition of decision tables and decision rules

– Active support

– Three “strategies”:

• Direct

• ‘Use scale orders’ (based on dominance)

• ‘Use weights’ (based on attributes’ weights)

– Validation:

• Consistency (based on dominance)

• Completeness (% determined function values)

– Principle:

• ‘The user is always right’ (but warned if considered to be in error)

Creation


Transparency: Representation and visualization of decision rules

Creation

CAR

PRICE

low

medium

high

unacc

acc

good

exc

TECH.CHAR.

bad

acc

good

exc

DEXi 10.7.14 Page 1

Decision rules PRICE TECH.CHAR. CAR 60% 40% 1 high * unacc2 * bad unacc3 medium acc acc4 medium good good5 low acc good6 >=medium exc exc7 low >=good exc

Aggregate

rules

3D point-by-point

graphic


Bridging the gap between qualitative and quantitative value functions

Creation

Principle

high

medium

low

badacc

good

exc

unacc

acc

good

exc

PRICETECH.CHAR.

CAR


Handling changes of model structure and components:

– Adding, deleting, moving, connecting attributes and subtrees

– Adding, deleting, moving, joining attribute values

Principles:

– Preserve the available information as much as possible

– Perform operations ‘behind the scene’ (with due warnings)

Creation


Handling changes of model structure and components:

– Adding, deleting, moving, connecting attributes and subtrees

– Adding, deleting, moving, joining attribute values

Example: delete attribute value ‘good’ of CAR

Creation


Evaluation of alternatives:

– Bottom-up table lookup

– Principle: Use all available information (even when data or rules are incomplete)

– Handling uncertainty:

• interval/set values

• probability distribution

• fuzzy distributions

Usage

Missing data Missing decision table COMFORT


Analysis of alternatives:

– “What-if analysis”

– “±1 analysis”

– Compare alternatives

– Selective explanation

Usage

OutlineOutline



• Methodology:





1. Decision Analysis and Decision Modelling1. Decision Analysis and Decision Modelling

Decision Analysis

Decision Makers+

Experts+

Decision Analysts

Decision/

Evaluation

model

decision

alternatives

A

B

C

D

E

Evaluation

Analysis

2. Data Mining and Decision Modelling2. Data Mining and Decision Modelling

Data

Data Mining

Decision

model

Decision Analysis

Decision Makers+

Experts+

Decision Analysts

Other sources

33. Decision Support Systems. Decision Support Systems

Data

Data Mining

Decision Analysis

Decision Support System Decision Makers+

Experts+

Decision Analysts

Other sources

Decision

model

JSI Collection of DEX ModelsJSI Collection of DEX Models

0

10

20

30

40

50

60

Nu

mb

er

of M

od

els

Model Timeline

RSRCH

EDU

DEMO

COM

Total: 104 projects, 582 models

What Kind of Projects?What Kind of Projects? • Computer Technology: software, hardware, IT tools, programming languages, DBMS, DSS, OCR

• Projects: investments, research, R&D, tenders

• Organisations: public enterprises, banks, business partners

• Schools: quality of schools, programmes and teachers, school admission, choosing sports

• Management: production, portfolio management, trade, personnel (employees, jobs, teams), privatization, motorway

• Production: location of facilities, technology, logistics, suppliers, office operations, construction, electric energy production, sustainability

• Ecology and Environment: dumpsite/deposit assessment and remediation, emissions, ecological impacts, soil quality, ecosystem, sustainable development, protected areas

• Medicine and Health Care: risk assessment (breast cancer, diabetes, ski injuries), nursing, technical analysis, knowledge management, healthcare network

• Agriculture and Food Production: economic and ecological effects of GMO, (un)approved GMO, crop protection, hop hybrids, garden quality

• Tourism: nature trail, tourism farm facilities, mountain huts

• Services: loans, housing loans, public portals, public services, leasing

• Other: cars, hotels, electric motors, radars, game devices, awards, roof covering, data mining

Medicine:Medicine:

Breast Cancer Risk AssessmentBreast Cancer Risk Assessment

Hormonalcircumstances

Personalcharacteristics Other

Menstrualcycle Fertility

Oralcontracept.

RISK

Cancerog.exposure

Fertilityduration

Reg. andstab. of men.

Age

First delivery

# deliveries

Quetel'sindex

Familyhistory

Demograph.circumstance

Physicalfactors

ChemicalfactorsMenopause

Bohanec, M., Zupan, B., Rajkovič, V.: Applications of qualitative multi-attribute decision models in health care, International Journal of Medical Informatics 58-59, 191-205, 2000.

Cropping Systems: Ecology PartCropping Systems: Ecology Part

Bohanec, M., Messéan, A., Scatasta, S., Angevin, F., Griffiths, B., Krogh, P.H., Žnidaršič, M., Džeroski, S.:

A qualitative multi-attribute model for economic and ecological assessment of genetically modified crops.

Ecological Modelling 215, 247-261, 2008.

CONTEXT CROP MANAGEMENT

soil state

nutrition state

CROP PROTECTION

weed control

pest control

disease control

weed profile

climate soil farm type chemical

fertiliz. use

soil tillage

water managmt

crop sub-type

biodiversity soil

biodiversity water quality

greenhouse gasses

ECOLOGY

herbivores

pollinators

weed biomass

predators parasitoids indirect

CO2 CO2 N2O

runoff water

undergrnd water

pesticide use

fertilizer use

fuel use

herbicide use

insecticideuse

fungicide use

physical stress

physical disturbance

climatic disturbance

soil fertilization

chemical disturbance

weed ctrl. applications

Traffic Control CenterTraffic Control Center

Omerčević, D., Zupančič, M., Bohanec, M., Kastelic, T.:

Intelligent response to highway traffic situations and road incidents.

Proc. TRA 2008, Transport Research Arena Europe 2008, 21-24 April 2008, Ljubljana.

Assessment of Reputation Risk in BanksAssessment of Reputation Risk in Banks FP7 ICT Large scale information extraction and integration

2010-2013 infrastructure for supporting financial decision making

PRODUCT

CUSTOMER

Ba

sic

data

pro

ce

ssin

g

Qualitative DEXi model

qRI1

Aggregation

RNP [40%]

RVP [60%]

→ Customer → Product → Counterpart → Bank

RI

COUNTERPART

bank data

RI: Reputation Index

RVP: Relative product volumes

RNP: Relative product numbers

Bohanec, M., Aprile, G., Costante, M., Foti, M., Trdin, N.:

A hierarchical multi-attribute model for bank reputational risk assessment.

DSS 2.0 - Supporting Decision Making with New Technologies (eds. Phillips-Wren, G., et al.),

Amsterdam: IOS Press, 92-103, 2014.

Ski Injury PredictionSki Injury Prediction

Bohanec, M., Delibašić, B.: Data-mining and expert models for predicting injury risk in ski resorts.

Decision Support Systems V - Big Data Analytics for Decision Making, First International Conference

ICDSST 2015, Belgrade, Serbia, May 27-29, 2015, Springer, 46-60, 2015.

Skiers

Crowding

numSkiers

numPasses

utilization

Weather

tempAvg

windSpeed

cloudiness

Skiing

TimeRun

avgAvgTimeRun

avgMinTimeRun

avgMaxTimeRun

Elevation

avgAvgElevation

avgMinElevation

avgMaxElevation

SpeedRunavgNoRuns

avgSpeedRun

Electric Energy Production TechnologiesElectric Energy Production Technologies

Technology

Rationality

Contribution to development

Economic

Societal

Economic-Technical advancementTechnical level

Expected development

Economy

Financial aspects

Energy price

Financing

Financial sources

Financial shares

Long-term liabilitiesEfficiency

Energy ratio

Return period

Independence Dependence

Land use and pollution

Spatial availability

Land availability

Energy share provision

Resource protection

Water protection

Land protection

Landscape protectionPollution

Health impactAir pollution

Greenhouse gases

Other pollutants

Public health status Contribution to development

Feasibility

Technical feasibility

Technological complexity

Infrastructure availability

AccessibilityFuel availability

Fuel accessibility

Economic feasibilityInvestment feasibility

Return of investment

Spatial feasibilitySocietal feasibility

Social acceptance

Permitting

Spatial suitability

UncertaintiesTechnological dependence Foreign dependence

Construction Licences

Operation

LicencesContracts

Special materialsWeather dependence

Fuel supply dependencePolitical stability

Possible changesPossible societal changes

Possible world changesPerception of risks

Kontić, B., Bohanec, M., Kontić, D.,

Trdin, N., Matko, M.: Improving appraisal

of sustainability of energy options - A

view from Slovenia, Energy Policy 90,

154-171, 2016.

Bohanec, M., Trdin, N., Kontić, B.: A

qualitative multi-criteria modelling

approach to the assessment of electric

energy production technologies in

Slovenia. Central European Journal of

Operations Research, 611-625, 2017.

Decision Support SystemDecision Support System http://sepo.ijs.si/naloge/OVJE/energetic_scenario_comparative_model/

http://sepo.ijs.si/naloge/OVJE/energetic_scenario_comparative_model/

http://sepo.ijs.si/naloge/OVJE/energetic_scenario_comparative_model/

GM_Presence

TraceabilityData

Products ProductGMPresence

CropGMPresence

GeoGMPresence

EU

GM_RegionProductComplexity

Countries

NumberCountries

CountryGMPresence

CoexistenceMeasures

Transportation

PrepackagedProduct

Logistics

LogComplexity

NumberInteractions

NumberCompanies

LogStorage

Harbour

Silo

SystemsUsed

TraceabilitySystemInPlace

IP_GMO

IP_Other

AnalCtrl_Systems

PrivateContracts

AnalyticalData

AnalyticalResults

AnalyticalResultsAvailable

ApprovedGMOsIdentified

UnapprovedGMOsIdentified

Methods

ProcessingLevel

AppropriateSampling

AppropriateMethods

Reliability

ReliabilityForApprovedGMO

RelevantGMCropsIncluded

AllIngredientsIncluded

OmnipresentGMOsIncluded

NumberScreenElem

ReliabilityForUnapprovedGMO

NumberScreenElem

AppropriateDataAnalysis

AppliedQualitySystem

ValidatedMethods

AccreditedLab

GMO Presence in Food and FeedGMO Presence in Food and Feed

INPUTS Product Data

OUTPUTS

Bohanec, M., Mileva-Boshkoska, B., Prins, T.W., Kok, E.: SIGMO: A decision support system for

identification of genetically modified food or feed products. Food Control, 71, 168-177, 2016.

DSS for the Assessment of GM DSS for the Assessment of GM Products Products http://decathlon.ijs.si/gmohttp://decathlon.ijs.si/gmo//

http://nejctrdin.com/ovjeGEN/

http://decathlon.ijs.si/gmo/

http://decathlon.ijs.si/gmo/

http://nejctrdin.com/ovjeGEN/

Parkinson’s Disease: Medication ChangeParkinson’s Disease: Medication Change

CarePlan

Motor

bradykinesia

tremor

gait

dyskinesia

on/off fluctuations

Non-Motor

daytime sleep.

cog.disorder

impulsivity

depression

hallucinations

Epidemiologic

age

employment

living alone

Structure

Decision rules

DEXi MedChange_Andrea.dxi 26.10.16 Page 1

Tables Motor Non-Motor Epidemiologic CarePlan 50% 50% 0% 1 problematic * * change2 <=maybe <=maybe * change3 * problematic * change4 maybe normal * maybe5 normal maybe * maybe6 normal normal * no_change bradykinesia tremor gait dyskinesia on/off fluctuations Epidemiologic Motor 19% 19% 24% 13% 15% 10% 1 problematic problematic * * * * problematic2 problematic * * <=problematic * * problematic3 problematic * * * problematic * problematic4 problematic * * * * active problematic5 * problematic * <=problematic * * problematic6 * problematic * * problematic * problematic7 * problematic * * * active problematic8 * * problematic * * * problematic9 * * * severe * * problematic

10 * * * * problematic active problematic11 problematic normal normal normal normal passive maybe12 normal problematic normal normal normal passive maybe13 normal normal normal problematic * passive maybe14 normal normal normal >=problematic problematic passive maybe15 normal normal normal problematic normal * maybe16 normal normal normal normal normal * normal daytime sleep. cog.disorder impulsivity depression hallucinations Epidemiologic Non-Motor 10% 10% 23% 23% 23% 10% 1 problematic * * * * active problematic2 * problematic * * * active problematic3 * * problematic * * * problematic4 * * * problematic * * problematic5 * * * * problematic * problematic6 problematic * normal normal normal passive maybe7 * problematic normal normal normal passive maybe8 normal normal normal normal normal * normal age employment living alone disease duration Epidemiologic 37% 38% 13% 13% 1 younger * * * active2 * employed * * active3 * * yes short active4 older unemployed * long passive5 older unemployed no * passive

DEXi MedChange_Andrea.dxi 26.10.16 Page 1

Tables Motor Non-Motor Epidemiologic CarePlan 50% 50% 0% 1 problematic * * change2 <=maybe <=maybe * change3 * problematic * change4 maybe normal * maybe5 normal maybe * maybe6 normal normal * no_change bradykinesia tremor gait dyskinesia on/off fluctuations Epidemiologic Motor 19% 19% 24% 13% 15% 10% 1 problematic problematic * * * * problematic2 problematic * * <=problematic * * problematic3 problematic * * * problematic * problematic4 problematic * * * * active problematic5 * problematic * <=problematic * * problematic6 * problematic * * problematic * problematic7 * problematic * * * active problematic8 * * problematic * * * problematic9 * * * severe * * problematic

10 * * * * problematic active problematic11 problematic normal normal normal normal passive maybe12 normal problematic normal normal normal passive maybe13 normal normal normal problematic * passive maybe14 normal normal normal >=problematic problematic passive maybe15 normal normal normal problematic normal * maybe16 normal normal normal normal normal * normal daytime sleep. cog.disorder impulsivity depression hallucinations Epidemiologic Non-Motor 10% 10% 23% 23% 23% 10% 1 problematic * * * * active problematic2 * problematic * * * active problematic3 * * problematic * * * problematic4 * * * problematic * * problematic5 * * * * problematic * problematic6 problematic * normal normal normal passive maybe7 * problematic normal normal normal passive maybe8 normal normal normal normal normal * normal age employment living alone disease duration Epidemiologic 37% 38% 13% 13% 1 younger * * * active2 * employed * * active3 * * yes short active4 older unemployed * long passive5 older unemployed no * passive

Heart Failure: Exercise PlanningHeart Failure: Exercise Planning Weekly Planning

DEXi EnduranceFrequency.dxi 15.5.17 Page 1

Scales Attribute Scale EnduranceFrequency 2x; 3x; 4x; 5x

Normative 2x; 3x; 4x; 5xCategory low; normalWeek 1; 2; 3; 4; 5; 6; 7; 8; 9; 10; 11; 12; 13; 14; 15; 16; 17; 18; 19; 20; 21; 22; 23; 24; more

Current 2x; 3x; 4x; 5xTransition decrease; stay; increase; automatic

MedicalAssessment decrease; stay; increase; automaticPatientsAssessment decrease; stay; increase; automatic

Category Week Normative

1 low <=4 2x

2 low 5–12 3x 3 normal <=6 3x

4 low 13–18 4x 5 normal 7–12 4x

6 low >=19 5x 7 normal >=13 5x

MedicalAssessment PatientsAssessment Transition

1 decrease * decrease 2 * decrease decrease

3 stay not decrease stay 4 not decrease stay stay

5 increase increase or automatic increase 6 increase or automatic increase increase

7 automatic automatic automatic

Pre-Exercise Medication Assessment

DEXi PreExerciseRequirements.dxi 16.5.17 Page 1

Scales Attribute Scale PreExerciseRequirements not_met; met

BloodCoagulationReasons yes; noTakesAnticoagulats yes; noPossibleBleeding yes; no

Rash yes; noHemorrhages yes; noNeurologicalSymptoms yes; no

MedicationIntakeReasons yes; noIntake<2hours yes; noExercisePreventionMedications yes; no

TakesBetaBlockers yes; noTakesACEInhibitors yes; noTakesARBs yes; noTakesDiuretics yes; noTakesLoopDiuretics yes; no

HeartRateReasons yes; noTakesDigitalis yes; noHR<45 yes; no

BloodPressureReasons yes; noHypertensionReasons yes; no

TakesACEInhibitors yes; noTakesARBs yes; noPersistentLowBloodPressure yes; noPersistentCough yes; no

SystolicPressureReasons yes; noTakesLoopDiuretics yes; noTookLoopDiuretics yes; noSYS<105 yes; no

SoftwareSoftware Personal standpoint: “Each MCDM method must be supported by

publicly available, user-friendly software”


Evaluators:

• JDEXi (java)

• DEXi.NET (C#)

• DEXiEval (command-line .exe)

Java Class Library:

• DEXx (incomplete)




OutlineOutline



• Methodology:





DEX: DEX: SummarySummary

• Suitable problems:

– Sorting/classification problems

– Difficult problems (many attributes and/or many alternatives)

– Problems that require human judgment, analysis, justification and explanation

– Problems with prevailing qualitative (rather than quantitative) indicators

– Finding solutions requires expert knowledge (decision rules)

– Recurrent decision problems (from decision to evaluation systems and DSS)

• Characteristics:

– Development of models: more engaging than ‘typical’ MCDA,

but still relatively simple and fast

– Qualitative models are less precise/discriminative than quantitative

(less suitable for choosing and ranking)

Future Plans and Future Plans and Research OpportunitiesResearch Opportunities

1. Extending the methodology

• Using numeric attributes: Combining qualitative and quantitative attributes

• Representing values with probabilistic and fuzzy distributions

– to cope with uncertainty both in alternatives and decision rules

• Relational models: to evaluate composed alternatives (e.g. company and departments)

Trdin, N., Bohanec, M.: Extending the multi-criteria decision making method

DEX with numeric attributes, value distributions and relational models. Central

European Journal of Operations Research, 1-24, 2017.

• Supporting dynamic aspects in extended models: e.g., acquisition, modification, analysis

• Hierarchical models (such as DEX, MCHP) as a platform, framework for other MCDM

methods


2. Software development

– DEXi software: Regular maintenance, no further extensions

– New generation DEX software:

• Goals:

– Extended DEX

– Using modern SW architectures and technologies

• Components:

– DEX Class Library

– Desktop application

– Web (client / server)

– unsure about mobile

– Software supporting individual or multiple MCDM methods

– Software platforms


3. Specific research topics and problems

A. Developing DEX models from data

– HINT (Hierarchy INduction Tool) (Zupan et al., 1998/99)

– proDEX (model revision) (Žnidaršič et al., 2004-06)

– New semi-automatic learning method:

Data + Model structure (+ Value scales) Decision tables

Idea and example in: (Bohanec & Delibašić, 2015)

Requires formalization and generalization

Data

Machine Learning



B. Ranking elementary decision rules &

Mapping qualitative to quantitative evaluations

– Method QQ (Bohanec, et al., 1992)

– Based on copulas (Mileva Boshkoska, et al., 2011-15)

– May be better formulated as an optimization problem

size age apartment uapartment

1 small old unacc 0,86

2 small medium unacc 1,00

3 small new unacc 1,14

4 medium old acc 2,00

5 medium medium good 2,81

6 medium new good 2,94

7 large old good 3,19

8 large medium exc 3,92

9 large new exc 4,08

premajhno

manjše

večje

staro

srednje

novo

nesprej

sprej

dob

odl

velikoststarost

stanovanje


Rules Weights • geometric

• information theoretic


C. Qualitative Quantitative transformations

Capacities

• integrals

• Shapley values

• interactions



D. DEX models with cycles

– Goal: Introducing loops in DEX models (e.g., ANP) dynamic models

– Preliminary thoughts and experiments: (Bohanec, et al., 2017)

Goal

Criterion 3Criterion 2Criterion 1

Goal

Criterion 3

Criterion 2

Criterion 1


4. Behavioral aspects of DEX modelling

5. Algorithms for “knowledge representation” (e.g., rules)

6. Search/optimization algorithms for decision analysis

(e.g., “option generator”)

7. Exploring special properties of decision tables (e.g., symmetricity)

8. Possible interactions:

– DEX : DRSA

– DEX : MCHP

9. Applications

10. Projects

Our Our ExperienceExperience in EU in EU ProjectsProjects

Logo Project Duration Description

Sol-Eu-Net 2000-2002 Data mining and decision support for business competitiveness

ECOGEN 2003-2006 Soil ecological and economic evaluation of genetically modified crops

SIGMEA 2004-2007 Sustainable introduction of genetically modified crops into European agriculture

Co-Extra 2006-2009 GM and non-GM supply chains: their CO-EXistence and TRAceability

HEALTHREATS 2007-2010 Integrated decision support system for health threats and crises management

FIRST 2010-2013 Large scale information extraction and integration infrastructure for supporting financial decision making

e-LICO 2010-2013 e-Laboratory for interdisciplinary collaborative research in data mining and data-intensive sciences

DECATHLON 2013-2016 Development of cost efficient advanced DNA-based methods for specific traceability issues and high level on-site applications

PD_manager 2015-2017 mHealth platform for Parkinson’s disease management

HeartMan 2016-2018 Personal decision support system for heart failure management

NARSIS 2017-2021 New Approach to Reactor Safety ImprovementS

Summary and Conclusion (1/2)Summary and Conclusion (1/2)

• DEX: – Multi-Attribute decision modeling methodology: hierarchical, qualitative, rule-

based

– A pioneering approach, combining multi-criteria decision modeling with rule-

based expert systems

• Contributions:

– Scientific, technical and practical

– Three generations of software: DECMAK, DEX, DEXi

– Hundreds of real-life applications

• Status:

– Conceived 30+ years ago, but alive: internationally recognized, actively used

in new projects, taught in schools, still developing

Summary and Conclusion (2/2)Summary and Conclusion (2/2)

• Important topics (too often overlooked): – Dynamic aspects of decision modeling

– Supporting software

• Future:

– Method extensions: added representations and features MCDM platform

– Implementation: extended, more powerful methodology on new architectures

– Further research:

fusion of methods, machine learning, rule ranking, dynamic DEX, ...

http://ifip2018dss.ijs.si/

http://ifip2018dss.ijs.si/

Thank youThank you

Marko Bohanec

[email protected]

method dex: a qualitative, hierarchical, rule-based...

Documents