method dex: a qualitative, hierarchical, rule-based...
TRANSCRIPT
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
Method for qualitative multi-attribute modeling
DEX is different from the majority of multi-attribute methods:
1. Attributes are discrete, symbolic, qualitative
CAR CAR
TECH.CH. TECH.CH. PRICE PRICE
COMFORT COMFORT SAFETY SAFETY MAINT MAINT BUYING BUYING FUEL FUEL
BUYING {high, medium, low}
FUEL {low, medium, high}
TECH.CH. {bad, acc, good, exc}
DEXDEX
Method for qualitative multi-attribute modeling
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
TECH.CH. TECH.CH. PRICE PRICE
COMFORT COMFORT SAFETY SAFETY MAINT MAINT BUYING BUYING FUEL FUEL
BUYING {high, medium, low}
FUEL {low, medium, high}
TECH.CH. {bad, acc, good, exc}
DEXDEX
Method for qualitative multi-attribute modeling
DEX is different from other multi-attribute methods:
2. Evaluation of alternatives (aggregation) is defined by decision tables
CAR CAR
TECH.CH. TECH.CH. PRICE PRICE
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
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
DEX Model: Formal RepresentationDEX Model: Formal Representation
DEX Model: Formal RepresentationDEX Model: Formal Representation
basic attributes - inputs
aggregate attributes
DEX Model: Formal RepresentationDEX Model: Formal Representation
basic attributes - inputs
aggregate attributes
outputs
DEX Model: Formal RepresentationDEX Model: Formal Representation
DEX Model: Formal RepresentationDEX Model: Formal Representation
DEX Model: Formal RepresentationDEX Model: Formal Representation
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
DEX Method: Dynamic AspectsDEX Method: Dynamic Aspects
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
DEX Method: Dynamic AspectsDEX Method: Dynamic Aspects
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
DEX Method: Dynamic AspectsDEX Method: Dynamic Aspects
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
DEX Method: Dynamic AspectsDEX Method: Dynamic Aspects
Bridging the gap between qualitative and quantitative value functions
Creation
Principle
high
medium
low
badacc
good
exc
unacc
acc
good
exc
PRICETECH.CHAR.
CAR
DEX Method: Dynamic AspectsDEX Method: Dynamic Aspects
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
DEX Method: Dynamic AspectsDEX Method: Dynamic Aspects
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
DEX Method: Dynamic AspectsDEX Method: Dynamic Aspects
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
DEX Method: Dynamic AspectsDEX Method: Dynamic Aspects
Analysis of alternatives:
– “What-if analysis”
– “±1 analysis”
– Compare alternatives
– Selective explanation
Usage
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
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/
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/
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”
http://kt.ijs.si/MarkoBohanec/dexi.html
Evaluators:
• JDEXi (java)
• DEXi.NET (C#)
• DEXiEval (command-line .exe)
Java Class Library:
• DEXx (incomplete)
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
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
Future Plans and Future Plans and Research OpportunitiesResearch Opportunities
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
Future Plans and Future Plans and Research OpportunitiesResearch Opportunities
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
Future Plans and Future Plans and Research OpportunitiesResearch Opportunities
3. Specific research topics and problems
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
Future Plans and Future Plans and Research OpportunitiesResearch Opportunities
Rules Weights • geometric
• information theoretic
3. Specific research topics and problems
C. Qualitative Quantitative transformations
Capacities
• integrals
• Shapley values
• interactions
Future Plans and Future Plans and Research OpportunitiesResearch Opportunities
3. Specific research topics and problems
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
Future Plans and Future Plans and Research OpportunitiesResearch Opportunities
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/