decision support method dex: concepts and...
TRANSCRIPT
Decision Support Method DEX: Concepts and Applications
Marko Bohanec
Jožef Stefan Institute, Department of Knowledge Technologies, Ljubljana, Slovenia and
University of Nova Gorica, Nova Gorica, Slovenia
FON, Belgrade, 13.12..2016
Contents
• Context: Multi-Criteria Decision Analysis (MCDA) • Method DEX (Decision EXpert)
– Method: Approach and Basic Concepts – Implementation: DEXi software – Two use cases:
• Job selection • Electric energy production technologies
• Outro: – Experience and other applications – Recent advances and future plans
What is MCDA, MCDM?
MCDA Multi Criteria Decision Analysis Multi Criteria Decision Aid(ing)
MCDM Multi Criteria Decision Making
Multi Criteria Decision Modelling
MADA, MADM Multi Attribute Decision ↑
A Typical MCDA Use Case
Multi-Criteria Model
evaluation
Alternatives
analysis
Performance variables
price
performance
quality
...
Decision Tasks (“Problematics”)
alternatives Choosing
✗
✗
✔
✗
✗
Sorting (Classification)
Ranking
Participants in MCDA
Multi-Criteria Model
evaluation
analysis
Performance variables
price
performance
quality
... decision makers
(decision owners) (stakeholders)
+ experts
+ decision analysts
x1
x2 f(x1,x2)
xn
y
Decision-Making Team
MCDA Methods
Wikipedia: https://en.wikipedia.org/wiki/Multiple-criteria_decision_analysis
`
Contents
• Context: Multi-Criteria Decision Analysis (MCDA) • Method DEX (Decision EXpert)
– Method: Approach and Basic Concepts – Implementation: DEXi software – Two use cases:
• Job selection • Electric energy production technologies
• Outro: – Experience and other applications – Recent advances and future plans
What is DEX? Decision EXpert Originates in 1980’s DEX
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
Fuzzy sets • verbal measures • fuzzy operators
DEX Method: History
1980 2000 1990 2010
DECMAK
Methodology • further improvement Software • DEXi Education International applications • Sol-Eu-Net • agriculture, food, GMO • project evaluation • finance 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
DEX Models by Time and Category
Total: 582 models
0
10
20
30
40
50
60
Num
ber o
f Mod
els
RSRCH
EDU
DEMO
COM
DEX Method for qualitative multi-attribute modeling DEX is similar to other multi-attribute methods:
1. Multiple attributes, hierarchically structured
2. Evaluation of alternatives: bottom-up aggregation
CAR
TECH.CH. PRICE
COMFORT SAFETY MAINT BUYING FUEL
Some Car
DEX Method for qualitative multi-attribute modeling DEX is different from the majority of multi-attribute methods:
1. Attributes are discrete, symbolic, qualitative
CAR
TECH.CH. PRICE
COMFORT SAFETY MAINT BUYING FUEL
BUYING ∈ {high, medium, low}
FUEL ∈ {low, medium, high}
TECH.CH. ∈ {bad, acc, good, exc}
DEX 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)
CAR
TECH.CH. PRICE
COMFORT SAFETY MAINT BUYING FUEL
BUYING ∈ {high, medium, low}
FUEL ∈ {low, medium, high}
TECH.CH. ∈ {bad, acc, good, exc}
DEX 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
TECH.CH. PRICE
COMFORT SAFETY
MAINT BUYING
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: Dynamic Aspects
How to: • Obtain model and its
components? • Verify model and its
components (e.g. for completeness and consistency)?
• Deal with uncertainty? • Ensure transparency,
comprehensibility? • Support model
dynamics?
Creation Usage
How to: • Obtain and represent
data about alternatives? • Deal with incomplete,
uncertain data? • Explain and justify
results? • Validate results? • Carry out the analyses?
Which analyses? • Assess the quality of
decision?
DEXi: Program for Multi-Attribute 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
A Didactic Use-Case: Job Selection
One Thursday morning, Charles, instead of attending his Management Science Techniques for Consultants class, was mulling over his four job offers. His offers came from: • Acme Manufacturing, • Bankers Bank, • Creative Consulting, and • Dynamic Decision Making. He knew that factors such as • location, • salary, • amount of management science (which he loved), and • long term prospects were important to him. He wanted to evaluate each job offer.
Adapted from: Michael A. Trick, Analytic Hierarchy Process, http://mat.gsia.cmu.edu/mstc/multiple/node4.html
DEX Method: Model Structure
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: Decision Rules
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: Representing Decision Rules
Transparency: Representation and visualization of decision rules
Creation
Aggregate rules
3D point-by-point graphic
MS long satisfaction 50% 50% 1 unacc * unacc2 * unacc unacc3 acc acc acc4 >=acc good good5 good >=acc good
satisfaction
MS
good
acc
unacc
unacc
acc
good
long
unacc
acc
good
DEX Method: Weights
Bridging the gap between qualitative and quantitative value functions
Creation
Principle
high
medium
low
badacc
goodexc
unacc
acc
good
exc
PRICETECH.CHAR.
CAR
DEX Method: Changing Model Structure
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: Changing Model Structure
Handling changes of model structure and components: – Adding, deleting, moving, connecting attributes and subtrees – Adding, deleting, moving, joining attribute values Example: add attribute value ‘exc’ to ‘long’
Creation
DEX Method: Evaluation
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
Complete data Incomplete data
DEX Method: Analyses
Analysis of alternatives: – “What-if analysis” – “±1 analysis” – Compare alternatives – Selective explanation
Usage
Contents
• Context: Multi-Criteria Decision Analysis (MCDA) • Method DEX (Decision EXpert)
– Method: Approach and Basic Concepts – Implementation: DEXi software – Two use cases:
• Job selection • Electric energy production technologies
• Outro: – Experience and other applications – Recent advances and future plans
Introduction to the Use-Case
• Electric energy production: – Is a strategic resource and activity for every country – Requires strategic management and planning years ahead – Is complex: technically, organisationally, financially, ... – Depends on needs, resources, technologies, ...
• Slovenia: – Currently no clear national strategy, slow progress – Ad-hoc and expensive (mis)investments – Competing approaches: “green”, “conventional” (coal, hydro,
gas), “nuclear” – Increasing energy demands
Introduction to the Use-Case
• Aims of this study (Project OVJE, 2013-14) – Identify reliable, rational, and environmentally sound
production of electric energy in Slovenia by 2050 – Consider technologies:
hydro, coal, oil, gas, nuclear, biomass, photovoltaic, wind – Assess individual technologies and technology mixtures
• Approach: – Multi-Criteria Decision Modelling (MCDM) – Qualitative MCDM method DEX (enhanced with uncertainty
intervals and probability distributions) – Assessing scenarios of technology mixture implementations
by 2050
Electricity Production in Slovenia • Annual consumption: 13 TWh • Production: approximately equal shares of
thermal, hydro, and nuclear power plants • Nuclear:
– Krško: 5 TWh/year until 2023 or 2043 • Coal-fired:
– old TEŠ 5: until 2027 – new TEŠ 6: 3.5 TWh/year
• Hydro: – all available resources used, – except lower Sava river
• Considering other technologies: – gas – wind, biomass, solar
Methods
Stage 1: Evaluation of individual technologies Hydro Nuclear Wind PV
Coal Gas Bio Oil
Model T “Technology”
Model M “Technology Mix”
Oil
Hydro
Coal Gas
Nuclear PV
Wind
Hydro
…
Stage 2: Evaluation of technology mixtures
Stage 3: Simulation of technology mixtures through time
2014 2050
Scenarios of power-plant utilization
Stage 1: Model T: Top Structure
Technology
Rationality
Contribution to development
Economy
Land use and pollution
Feasibility
Technical feasibility
Economic feasibility
Spatial feasibility
Uncertainties
Technological dependence
Possible changes
Perception of risks
Stage 1: Model T: Full Structure
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 feasibility Societal feasibilitySocial acceptance
Permitting
Spatial suitability
UncertaintiesTechnological dependence Foreign dependence
Construction Licences
Operation
LicencesContracts
Special materialsWeather dependence
Fuel supply dependencePolitical stabilityPossible changesPossible societal changes
Possible world changesPerception of risks
Total 66 attributes: 36 basic 28 aggregate 2 link
Stage 1: Model T: Decision Rules
Land use and pollution unsuit; less_suit; suit; more_suitSpatial availability less_suit; suit; more_suit
Land availability low; med; highEnergy share provision low; med; highResource protection weak; present; effective
Water protection weak; present; effectiveLand protection weak; present; effectiveLandscape protection weak; present; effective
Pollution high; med; lowHealth impact high; med; low
Air pollution high; med; lowGreenhouse gases high; med; lowOther pollutants high; med; low
Public health status low; med; highContribution to development low; med; high
Rules are checked for: • completeness: coverage of all possible cases • consistency: obeying the principle of dominance
Stage 1: Model T: Decision Rules TechnologyFeasibility=med
Rationality
high
med
low
inapprop
unsuit
weak
suit
good
exc
Uncertainties
v_high
high
med
low
none
Rationality Feasibility Uncertainties Technology 1 inapprop * * unsuit2 <=low <=med v_high unsuit3 <=med low v_high unsuit4 >=low low high:med weak5 >=low high v_high weak6 >=med >=med v_high weak7 high low <=med weak8 high * v_high weak9 low:med low >=low suit
10 >=low low low suit11 >=low >=med high suit12 low >=med >=med good13 low:med med med:low good14 >=low >=med med good15 high low none good16 >=med >=med none exc17 >=med high >=low exc18 high >=med >=low exc
1. Model T: Results Attribute Hydro Coal Oil Gas Nuclear Bio PV Wind Technology suit - exc unsuit unsuit weak - good weak - exc unsuit unsuit unsuit Rationality low - high inapprop inapprop high high inapprop inapprop - low inapprop Contribution to development med - high high med high high med low - med low Economy med - high high low med - high med - high low low low Land use and pollution less - more_suit unsuit unsuit more_suit more_suit less_suit unsuit - more_s. unsuit - less_suit Feasibility high high high high low - high low - med low low Technical feasibility high high high high high med med - high med Economic feasibility high med med med high low - med low low Spatial feasibility high high high high low - high low - high low - high low - high Uncertainties high - none low v_high - low v_high - med v_high - low low v_high v_high Technological dependence high - none low v_high - med v_high - med v_high - low med v_high v_high Possible changes pos pos no no pos no no no Perception of risks med - none med - low none high - med v_high - low none low none
Stage 1: Model T: Results Attribute Hydro Coal Oil Gas Nuclear Bio PV Wind Technology suit - exc unsuit unsuit weak - good weak - exc unsuit unsuit unsuit Rationality low - high inapprop inapprop high high inapprop inapprop - low inapprop Contribution to development med - high high med high high med low - med low Economic high high med - high high high med low low Societal med - high high med high high med low - med low Economic-Technical advancement med high med med - high high med high high Technical level med high med high high med high high Expected development low - med med - high low - med low - med high low med - high high Economy med - high high low med - high med - high low low low Financial aspects more_suit suit - more_suit less_suit suit more_suit less_suit less_suit less_suit Energy price med - low high - med high med low high high high Financing more_suit more_suit suit suit more_suit less_suit less_suit less_suit Financial sources certain certain less_certain certain certain uncert - less_cert less_cert - certain less_cert - certain Financial shares more_suit more_suit suit suit more_suit less_suit less_suit less_suit Long-term liabilities more_suit suit suit less_suit suit less_suit less_suit less_suit Efficiency med - high high med high high med low low - med Energy ratio med - high high med high high med low low Return period med - short short med med - short short long - med long med - short Independence low - high high low - med low - med low - high med low low Dependence high - none low v_high - med v_high - med v_high - low med v_high v_high Land use and pollution less - more_suit unsuit unsuit more_suit more_suit less_suit unsuit - more_s. unsuit - less_suit Spatial availability less_suit - suit less_suit - suit suit - more_suit more_suit more_suit less_suit less_suit - suit less_suit - suit Land availability med - high med - high high high high low med low - med Energy share provision low med low - med med - high high low low low Resource protection weak - effective weak - present effective effective effective weak - present weak - present weak - present Water protection weak - effective weak - present present effective present present effective effective Land protection present - effective weak - present effective effective effective weak - present weak - present weak - present Landscape protection present - effective weak - present effective effective effective weak - present weak - present weak - present Pollution low high med low low med low low Health impact low med high med low med high - low high Air pollution low high high high low med low low Greenhouse gases low high high high low low low low Other pollutants low high med med low med low low Public health status med - high high med high high med low - med low Contribution to development med - high high med high high med low - med low Feasibility high high high high low - high low - med low low Technical feasibility high high high high high med med - high med Technological complexity more_suit more_suit suit suit more_suit suit less_suit - suit less_suit - suit Infrastructure availability high high high high high low - med low - med low Accessibility high high med med high med high high Fuel availability high high med med high high high high Fuel accessibility high high med med high low high high Economic feasibility high med med med high low - med low low Investment feasibility high med med med high low - med low low Return of investment suit - more_suit suit - more_suit suit suit suit - more_suit suit less_suit - suit less_suit - suit Spatial feasibility high high high high low - high low - high low - high low - high Societal feasibility med - high med - high med med - high low - high low - med low - high low - high Social acceptance med - high med - high med med - high low - high low - med low - high low - high Permitting yes yes yes yes yes yes yes yes Spatial suitability high high high high high high high high Uncertainties high - none low v_high - low v_high - med v_high - low low v_high v_high Technological dependence high - none low v_high - med v_high - med v_high - low med v_high v_high Foreign dependence high - none low v_high - med v_high - med v_high - low med v_high v_high Construction low med med med high - med med med med Licences no restr moder_restr moder_restr moder_restr strong - moder moder_restr moder_restr moder_restr Operation high - low low high - med high - med high - low med high high Licences no restr moder_restr moder_restr moder_restr strong - moder moder_restr moder_restr moder_restr Contracts no restr moder_restr moder_restr moder_restr strong - moder moder_restr moder_restr moder_restr Special materials no restr moder – no moder_- no moder_ - no strong - moder moder_- no moder_restr moder_restr Weather dependence high - med low low low low med high high Fuel supply dependence med - low low high - med high - med high - low med - low low low Political stability high high low low high high high high Possible changes pos pos no no pos no no no Possible societal changes pos pos no no pos no no no Possible world changes no pos no no no no no no Perception of risks med - none med - low none high - med v_high - low none low none
Methods
Stage 1: Evaluation of individual technologies Hydro Nuclear Wind PV
Coal Gas Bio Oil
Model T “Technology”
Model M “Technology Mix”
Oil
Hydro
Coal Gas
Nuclear PV
Wind
Hydro
…
Stage 2: Evaluation of technology mixtures
Stage 3: Simulation of technology mixtures through time
2014 2050
Scenarios of power-plant utilization
Stage 3: Evaluation of Scenarios Scenario: Some realization of management decisions in 2014-2050 Management decisions: 1. Closing-down of the nuclear power plant (NPP) Krško Unit1 in 2023. 2. Construction of Unit2 at the NPP Krško by 2025. 3. Finalisation of the two hydro power plants on the lower Sava river by 2025. 4. Construction of a gas fired power plant by 2025. 5. Closing-down of Unit5 of the coal fired power plant at Šoštanj in 2027. 6. Construction of the chain of hydro power plants on the mid Sava river by 2035.
Implementation http://sepo.ijs.si/naloge/OVJE/energetic_scenario_comparative_model/
Results • Contributions:
– Systematic, transparent and reproducible sustainability appraisal of technologies and strategic management scenarios for Slovenia
– Qualitative MCDM approach: DEX Models T and M – Assessment of technology mixtures in time rather than only individual
technologies – Decision support system for evaluation of scenarios
• Findings and recommendations:
– Strategic planning of electric energy production in Slovenia is already very late – Three most suitable technologies for Slovenia: Hydro, Gas, and Nuclear – Only a proper mixture of these technologies is reliable and rational for meeting
expected energy needs – Biomass, wind and photovoltaic sources of energy are less suitable than others
and may provide only from 8% to 15% of energy in Slovenia
Further Information
Contents
• Context: Multi-Criteria Decision Analysis (MCDA) • Method DEX (Decision EXpert)
– Method: Approach and Basic Concepts – Implementation: DEXi software – Two use cases:
• Job selection • Electric energy production technologies
• Outro: – Experience and other applications – Recent advances and future plans
Decison Problems Addressed by DEX • 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, options, drug addiction, roof
covering, data mining
Medicine: Breast 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.
Ski Injury Risk Assessment
Bohanec, M., Delibašić, B.: Data-mining and expert models for predicting injury risk in ski resorts, ICDSST 2015, Belgrade, Serbia.
Crowding Weather Skiers 1 H H H2 H M H3 H L M4 M H H5 M M M6 M L L7 L H M8 L M M9 L L L
tempAvg windSpeed cloudiness Weather 1 L * * H2 * L L H3 >=M L >=M M4 >=M >=M * L
noSkiers noPasses utilization Crowding 1 H H <=M H2 H * H H3 * H H H4 H * L M5 <=M <=M L M6 * H L M7 H >=M >=M M8 <=M M >=M M9 M <=M >=M M
10 >=M H >=M M11 M M * M12 >=M >=M H M13 >=M L >=M L14 L >=M >=M L
Skiers
Crowding
numSkiers
numPasses
utilization
Weather
tempAvg
windSpeed
cloudiness
Cropping 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 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 GMOs in Food and Feed
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.
2013-2016 DECATHLON FP7-KBBE-613908: Development of cost efficient advanced DNA-based methods for specific traceability issues and high level on-site applications
Context: • Genetically modified organisms (GMOs) • GMO trade and marketing is strictly regulated in Europe • In the world, the situation is becoming increasingly complex • Difficult assessment of supply chains for potential presence of authorised and unauthorised GMOs Given: • Some food or feed product (raw or processed, bulk or packaged, simple or compound, ...) • Traceability data: country of origin, transportation data • [Analytical data: established presence of GMOs in the product] Goals: • Assess the likelihood of authorised or unauthorised GMOs presence in a given product • Provide a Decision Support System (DSS) for producers and traders
Assessment of GMO in Food and Feed
GM presence
TraceabilityData
AnalyticalData
Products
Countries
Transportation
ProductRisk
ProductComplexity
CropRisk
EU
GM_Region
NumberCountries
CountriesAtRisk
CoexistenceMeasures
PrepackedProduct
Logistics
SystemsUsed
LogComplexityNumberInteractions
NumberCompanies
LogStorage
Harbour
Silo
AppropriateSampling
RelevantGMCropsIncluded
NumberScreenElem
ValidatedMethods
AnalyticalResults
ProcessingLevel
GeoRisk
TraceabilitySystemInPlace
IP_GMO
IP_Other
AnalCtrl_Systems
PrivateContracts
AppropriateDataAnalysis
AccreditedLab
ApprovedGMOsIdentified
UnapprovedGMOsIdentified
AnalyticalResultsAvailable
Methods Reliability
ReliabilityForApprovedGMO
ReliabilityForUnapprovedGMO
AppropriateMethods
AppliedQualitySystem
AllIngredientsIncluded
OmnipresentGMIncluded
INPUTS Product Data
OUTPUTS
Assessment of GMO in Food and Feed
GM presence
TraceabilityData
AnalyticalData
Products
Countries
Transportation
ProductRisk
ProductComplexity
CropRisk
_
NumberCountries
CountriesAtRisk
CoexistenceMeasures
PrepackedProduct
Logistics
SystemsUsed
LogComplexity
LogStorage
AppropriateSampling
AnalyticalResults
ProcessingLevel
GeoRisk
TraceabilitySystemInPlace
_
IP_Other
_
PrivateContracts
ApprovedGMOsIdentified
UnapprovedGMOsIdentified
AnalyticalResultsAvailable
Methods Reliability
AppropriateMethods
AppliedQualitySystem
CropRisk GeoRisk ProductRisk 1 high <=med high2 <=med high high3 high low med4 med med med5 low high med6 >=med low low7 low >=med low
NumberCountries CountriesAtRisk CoexistenceMeasures Countries 1 * yes not all high2 >2 yes all med3 * no * low4 1-2 * all low
TraceabilityData AnalyticalData GM_Presence 1 v-high <=high v-high2 * high v-high3 <=high med high4 high no_data high5 med no_data med6 med med med7 <=low low low8 low no_data low9 low med:low low
10 * v-low v-low11 v-low no_data v-low12 v-low >=med v-low
DEX: Experience • 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) – Uncertainty (incomplete knowledge, imprecise or missing data) – 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) – Decision rules are “shallow”
DEX: Recent Advances
• 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 alternatives composed of sub-components
(e.g. company and departments)
Currently implemented in DEXx java library: https://bitbucket.org/nejctrdin/dexx
DEX: Recent Advances • DEXi HTML Evaluator: Running DEXi models in Web browsers
http://kt.ijs.si/MarkoBohanec/dexihtml.html
DEX: Future Plans • Implementation:
– DEXi software: Regular maintenance, no further extensions – DEX2: User-friendly implementation of the extended (DEXx) method – Implementation on new architectures: software library, desktop, Web, mobile
• Research and Development: – “Dynamic” DEX models: Models with cycles, similar to ANP – Semi-automatic development (“learning”) of DEX models from data – Behavioral aspects of developing DEX models – Integration/combination with other methods (AHP/ANP, DRSA, ROR, UTA*) – A number of specific issues:
• representations and visualizations of decision tables and rules • improved ranking of alternatives • taking advantage of value function properties, e.g. symmetricity • approximating value functions and assessing “local” weights • ...
Summary and Conclusion • 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 • Future:
– DEXi software: maintenance – Implementation: extended, more powerful methodology on new architectures – Further development: dynamic DEX, machine learning, specific improvements