Report Model-Based Decision support

1
system

合肥工业大学管理学院

智能决策支持系统

姓名 : 杜瑞

学号 : 2017010301

专业 : 管理科学与工程

班级 : 2018-2019

INTRODUCTION
 Report Model-Based Decision support
2
system

The field of decision support systems is one of the most enduring in the information systems
domain, having emerged in the 1960s, at the beginning of the history of information systems
as an area of research, from the pioneering work of Simon, Keen, and Scott Morton amongst
many others. Through five decades, this field has continued to attract considerable and
increasing attention from researchers and practitioners, under a range of banners.
Back in 1991, Teng and Galletta (1991) conducted a survey of IS researchers which showed
32% of respondents listed DSS as their area of research, with a further 22% listing artificial
intelligence and 21% database management systems. These three areas, which totalled 281
mentions out of 845 valid responses, were actually the top 3 IS research areas listed by Teng
and Galletta. In more recent times, Arnott and Pervan’s (2006) comprehensive review of the
DSS field yielded 1093 research papers published from 1990 to 2004 in 14 major journals, the
key journals in a broadly defined IS research area. These papers represented 15.2% of all
papers published in the 14 journals.
These statistics present a view of DSS as a core area of the IS domain and one with a
dynamism well illustrated by the strength of such international groupings as the AIS Special
Interest Group on Decision Support, Knowledge and Data Management Systems (SIG DSS),
the Working Group 8.3 of the International Federation for Information Processing (IFIP) on
DSS and the EURO Working Group on DSS (EWG DSS). Needless to say, researchers from
these groupings form a large proportion of the contributors to this encyclopedia. Nertheless,
as evidenced by the wide scope of the contibutions presented in the encyclopaedia, and the
range of backgrounds and expertise of their authors, the area of DSS can now successfully
claim to integrate groundings from a wide range of disciplines, ranging from Operations
Research, to Management and to Social Psychology, with excellent, insighful and sometimes
fascinating results. Notwithstanding the debates about the best vehicle to study the way
organisations work and what kind of systems to develop to support their needs, as illustrated
in Morgan’s (1986) superb Images of Organisations, we stand firm in our belief that thinking
of organisations as decision making entities and the systems to support
them as decision support system is a perspective that has served our discipline very well.
Though no perspective should ever be adopted in an exclusive fashion and to the detriment of
other equally valid perspectives, we will continue to preach the wide adoption of decisions
and decision support artefacts as objects of study, now and in the future. In passing, we will
smile knowingly when colleagues greet us with their usual joke: “But DSS
 Report Model-Based Decision support
3
system

is dead at the stage!” or “DSS! It’s just like flogging a dead horse!!” As this encyclopaedia
illustrates, the DSS horse is not only alive and well, it is prancing around the field; or has it
even jumped the fence to conquer new pastures. However, the success of the DSS field is also
one of its weaknesses: as the area has grown and knowledge from an increasing number of
domains is brought to bear on the problems that must be studied, it becomes increasingly
difficult to maintain a complete vision of DSS and all its latest developments. Already, it is
arguably the case that a researcher would struggle to keep a handle on all requisite knowledge
to be able to understand or review all the DSSs papers produced in a year. For instance, it
becomes difficult to keep pace with developments in Multi-Criteria Decision Making or in the
modelling of uncertainty on the one hand, as well as with our xxxi increased understanding of
the role of emotions in human decision making or with the lastest ideas of how to study
Recognition-Primed Decision Making. Furthermore, for students of DSS, young researchers
and practitioners who want to train themselves to be worthy of the DSS specialists label, it is
increasingly difficult to know what to learn and what skills to practice. This encyclopaedia
project is born out of these observations that the diversity in our domain requires a source
of information that does justice to the growth of DSS in recent years, offers the theoretical
basis for researching and developing DSSs and provides visibility on the topics where
research should be directed in the future. Of course, no single book can meet all the
requirements for such an ambitious a target, but we hope that this encyclopaedia of Decision
Making and Decision Support Technologies will contribute to the formalisation of the DSS
area in its current, diversified form. Repository of formal DSS knowledge, reference for the
research carried out to date, instructional manual, springboard for future research, this book
should help students, researchers, consultants and managers alike in their learning and
experimentation with DSS concepts and artefacts. In sourcing the material hereby presented,
we have appealed to a broad range of constituencies of researchers and practitioners
worldwide both within the specialised groups which explicitly work on DSS and in the wider
IS community. Response was excellent from this all quarters. This led to the submission of
over 200 proposals from which, after a lengthy reviewing process, 110 entries were selected
and finalised. In classifying the material and to facilitate browsing through the 110 entries, we
have relied on the authors of the contributions to provide each a keyword which best captured
the essence of their contributions.These keywords were used to create the thematic table of
contents (see Contents by Keyword) which was added to this encyclopedia in addition to the
 Report Model-Based Decision support
4
system

alphabetical table of contents of the paper titles. It is proposed that this provides a
straightforward way to navigate the 1000 or so pages of the two volumes of this encyclopedia,
either in leisurely fashion, following no particular sequence, or in a more orderly fashion, to
meet a specific research, learning or teaching need. The Key Terms listed at the end of each
contribution, compiled by each contributor, should also be an extremely valuable resource as
they represent over a thousand terms and definitions pertaining to our discipline which
students may be finding difficult to grasp and for which there may be no obvious source of a
recent definition. Definitions, of course, can be difficult to agree upon, but the whole point of
this broadly contributed compilation effort is precisely to offer a broad spectrum of
perspectives on DSS topics, such that the glossary featuring in these pages should be a true
representation of what DSS researchers around the world think is important in the area.
In concluding this preface, let us wish that these volumes represent an authentic, fresh and
useful book (in the words of one of our editorial board members) that adds value to the
domain, solves the problem of finding clear sources of information for the DSS researchers of
the future and provides fresh impetus to pursue the study and development of DSS artefacts
that support people and organisations as they make their decisions.
Decision support systems (DSS) are a subset of computer based information systems (CBIS)

There are a number of decision support systems, these can be categorized in five types of DSS
which are: Data-driven DSS, model driven DSS, knowledge driven DSS, document driven
DSS, and communication –driven and group DSS.

Model driven DSS includes systems that are used accounting and financial models
representational models and optimization models.

Research has demonstrated and substantiated that the first goal of DSS is to reduce decision
cycle time, increased employee’s productivity and more timely information for decision
making.

The purpose of our paper is to dissect the model-based decision support’s concept. Firstly we
are going to define this concept.

Some authors as Forgionne and AL quoted: A model-based DSS is “an interactive computer-
based system composed of a user-dialog system, a model processor and a data management
system, which helps decision makers utilize data and quantitative models to solve semi-
structured problems” (Forgionne et al., 2005, p. 765)

BACKGROUND AND MOTIVATIONS

 Report Model-Based Decision support
5
system

This part is to see decision’s process from the statement of the problem, through the necessary
tools that will be used until the resolution of the problem.
The characteristics of a decision process have always received most attention from social
scientists and economists, as this aspect is most directly related to the way decision making is
organized and to the process of choice. It is also surprising how decision situations can be
cleared up by structuring the decision process and clarifying the preferences. In several cases
such a clearing up does the job and is sufficient to understand which decisions should be
taken. Since the decision process has so many features, there are many ways of structuring the
process and clarifying the preferences. Therefore, a whole phalanx of methods and
approaches has been developed, mainly concentrating on one or two particular features.
However, the integration of various approaches for different features, in a case when more
features are relevant, is often quite straightforward.
Therefore, it will be sufficient for the moment to mention some features, together
with corresponding approaches. A typical feature of complex decisions is that they consist of
several related subdecisions. A typical relation is that one subdecision restricts the set of
alternatives for another. For instance, if I am planning my holidays, then I have to decide on
the destination, time, mode of travel, and type of accommodation. However, as soon as I
decide to go to Paris, the possible set of travel arrangements is restricted considerably. For
this feature the use of a decision tree can be very helpful. In such a tree, the subdecisions are
represented in a sensible order and in a well-structured way. This relatively simple tool may
help considerably; quite often, subdecisions become rather simple and apparent in this way.
Another important feature of decisions might be that they are not made by one person but by a
group of persons who have to agree somehow on the ultimate choice. There are various ways
of facilitating and supporting group decision making. Typically, they rest on two factors. The
first involves extracting from the group their priorities and the criteria that are relevant for the
decision. The second involves providing a means for discussing and exchanging evaluations
of possible alternatives according to the chosen criteria. A typical example is the choice by a
board of a new chief executive officer from a set of applicants provided by a headhunter. For
several other examples, the group feature has to be treated in an integrated way with other
features, for instance, with subdecisions. This occurs when I am planning holidays not just for
myself, but also for my family. Still another feature may be that there is some division of
power, causing several parties to make their own decisions, but with all parties feeling the
consequences of the decisions of the others. If one considers, for instance, the acid rain
problem in Europe, then one recognizes easily that each country may decide autonomously on
its emission for, say, the next five years, by issuing regulations on power generation and use,
on manure production and handling, and car driving. However, it is also clear that several
European countries suffer mostly from acid rain caused by the emissions of other countries.
Therefore, the strongest stimulus for a country to decrease its emissions may well be the
promise from neighboring countries that they will decrease their emissions as well. Clearly, in
this situation, the decisions should be based on negotiations of some sort. Hence the question
arises how one may facilitate or support negotiations. Negotiations support is even more
difficult than group decision support. In the case of negotiation each party has to decide for
itself upon the criteria and their relative relevance. There are many ways to achieve a
 Report Model-Based Decision support
6
system

conclusive decision. A way that is most appropriate in a particular situation depends very
much on circumstances. The last feature to mention here (although there are still many others)
is the phase of choice in the decision process. This is usually based on complicated
preferences for different alternatives. It has already been indicated (e.g., for the case of group
decision making) that there might be several relevant criteria. In the case of selecting a chief
executive officer, important criteria might be: (i) experience in similar functions; (ii) an
innovation-oriented personality; (iii) management quality;and (iv) financial expertise. Unlike
the single criterion in formulation for the first aspect, alternatives are not necessarily
evaluated with real numbers. Evaluation might equally be of a qualitative nature, such as very
high (v), high (h), and moderate (m) for management quality. These evaluations may even be
more complex if a criterion is subdivided into subcriteria; for instance, management quality
might be subdivided into analytic power and executive power. In the case of two such
subcriteria, one could have the following qualifications:
(v; v) (v; h) (v;m) (h; v) (h; h) (h;m) (m; v) (m; h) (m;m)
It is possible that the desirability of such qualifications cannot even be ordered in a simple
way. In the above-quoted example of management quality, it is clear that a person with (v; v)
is most desirable, but (m;m) is not necessarily worse than (m; v), as someone without good
ideas might be dangerous if she/he scores high on executive power. One of the tools for
making preferences operable is the multi-attribute utility theory (MAUT; Rios, 1994).
Another approach, based on slightly different assumptions, is the analytic hierarchy process
(AHP; Saaty, 1980). There are also other approaches, such as outranking relations (Vincke,
1992). Indeed, clarifying the preference structure in one way or another might be very helpful.
We can see that when analyzing the decision process, we actually build another part and type
of the model of the problem. We shall use the following basic distinction of model types:

 Substantive models, describing the relation between basic processes and actions and
trying to represent a part of reality relevant for the decision situation being analyzed.
 Preferential models, representing the preferences of the decision maker and
other aspects of the decision process.

OVERVIEW ON DECISION SUPPORT SYSTEMS

Model-based decision support systems

 Report Model-Based Decision support
7
system

Decision support systems cover a wide variety of systems, tools and technologies for
informing and supporting decision makers. Based on the tool or component that provides the
dominant functionality in a DSS, Power proposes a taxonomy of decision support systems by
distinguishing communication-based, data-based, document-based, knowledge-based, and
model-based DSS. Model-based decision support systems integrate different kinds of
mathematical and analytical models for simulation and prediction of the system1 behavior.
Hence, model-based DSS exploit the full resolution and detail of simulation models, thus
avoiding the pitfalls and limitations of the approximations often used for optimization. Key
issues when designing a model-based DSS are the choice of appropriate models and software,
and the definition of data formats. Very large databases are usually not needed for model-
based decision support systems.

DSS terminology and logic

It is assumed that the decision problem is structured and presented in terms of:
• The control options, i.e. the alternative actions, strategies, and/or policies that can be
undertaken to affect the system behavior.
• The criteria on whose basis the system performance led by each control option is evaluated.
• The objectives, i.e. the type of optimization to be performed on each criterion.
• The constraints, establishing bounds for some/all the criteria in order to make the evaluated
alternative acceptable or feasible. The aim of the DSS is to support the choice of a control
option that both meets the constraints and optimizes the objectives. In this respect, the control
option definition and design is of central importance. The control options are described by
value assignments of the controllable variables. Example: If local authorities are asked to
grant new farming concessions for aquaculture, and have to decide the amount of such
concessions, the allocated farming area is the controllable variable. On the other hand, the
uncontrollable variables describe external factors that are not subject to choice, but do affect
the system performance. Their role in the DSS can be only viewed in terms of sensitivity and
robustness of the final decision. Example. Typical uncontrollable variables are the weather
conditions and the water inflows for the biogeochemical models of a lagoon, and the prices
and market data for the economic models. The criteria are expressed by means of indicators,
which are used to synthesize the system performance under alternative control options. The
objectives correspond to indicators whose value has to be either maximized or minimized.
The constraints impose a maximum and/or minimum value to the indicators. They may
correspond to thresholds defined on the basis of regulations and/or experience, and allow to

discard unacceptable alternatives. Additional variables that do not correspond to criteria, but
the decision maker might want to constrain, are referred to as internal variables. Note that
 Report Model-Based Decision support
8
system

also the controllable and uncontrollable variables could be bounded in order to reflect both
physical and practical constraints. With the above definitions, the basic DSS logic is simple.
A set of alternative control options for the system are generated by changing the values of the
controllable variables. Each control option leads to a corresponding system performance,
which is synthesized by indicators. Performances are analyzed, evaluated and compared by
means of suitable multicriteria analysis tools to arrive at the final preference ranking of the
alternatives, and the eventual choice of a preferred alternative as the solution of the decision
process.
A bit higher we said that some typical approaches were sketched for each of the three aspects
of decision making. The present section is devoted to a discussion of those approaches. The
essential issue is, of course, the usefulness of a given approach. However, we know already
that each of the approaches can be very useful. Hence the remaining problem is the reverse
question: If we have a specific decision problem, which one of these approaches is then most
useful?
A satisfying answer to this question would require a classification of decision problems with
an allocation of one or more approaches to each class. Without trying to make such a
classification explicitly, one may already deduce some conclusions. It has already been
indicated that although the information aspect is quite basic, a concentration on information
alone might be insufficient for decision support. For instance, a management information
system may give all the information relevant to the production process of a firm, but it cannot
predict the consequences of a decision to halve delivery times. However, it may well
represent the financial and performance consequences of a structural change in the
distribution network. Several practical decision problems have certain features that make the
approach based on modeling natural and relevant. In particular, formulation or some of its
extended, less strict variants – typically involving many objectives – frequently provide the
natural setting. This is particularly the case if some relatively hard technical, physical, or
economic processes play the central role in the sense that they have to be influenced somehow
by the decision. However, in most cases that fit into this category, the decision process and
the informational structure are rather complicated. Consider, for instance, the acid rain
problem in Europe. The core processes consist of emissions (caused by technical, economic,
and social phenomena), and dispersion and deposition somewhere in Europe (here the
reigning winds and other climate conditions are decisive). Essentially, the aim is to change the
results of these core processes, and that can only be done by influencing the emissions, since
changing the wind directions is not feasible. However, the emissions themselves cannot be
influenced directly, but the processes that generate them may. Processes to be influenced
might include the mobility of the population, technical performance of cars, land use, ways of
feeding cattle, use of electricity, and ways of generating electricity. Information regarding
such processes can be well represented by data bases and other information systems.
 Report Model-Based Decision support
9
system

This discussion reveals that it is a complicated decision process to achieve a decrease in

emissions within any of the countries. It is already complicated to discover what measures
might be taken to achieve a particular decrease, but to agree on such measures is a completely
different issue. Thus, the core processes may be located easily in this case, but this does not
imply that they represent the core of the decision problem. With this typical case in mind, we
may conclude that a straightforward application of an approach based on the relation between
basic processes and actions is useful only for a small percentage of decision problems. There
might also be problems that represent the third aspect – the specificity of the decision process
– in almost pure form. For example, the partitioning of an inheritance among the children of a
deceased person might be an example of a negotiation problem in pure form (Raiffa, 1980).
Other problems might be, however, less purely negotiational. For example, the acid rain
problem also has this negotiation aspect, but the complex relation between decisions and
consequences plays a dominant role in this case. Generally, only a small minority of decision
problems can be treated well by selecting one aspect-oriented approach. We need approaches
that integrate more than one aspect. In this book we will show that such an integration
approach is possible for some institutional decision problems, for which the decision making
ultimately relies on influencing relatively hard basic processes. In particular, the approach
will be illustrated with environmental problems that belong to that category. This implies that
models of the basic processes and decisions should play a core role in our approach (without
denying the importance of the other aspects). Therefore, we will first comment on the relation
between basic processes and actions.

MATHEMATICAL PROGRAMMING FOR DECISION SUPPORT

Here we will have to discover that some various types of model-based decision support will
be outlined by presenting different types of models. This section will be devoted to models
that fit into formulation (2), with the exception that models with multiple criteria will also be
treated. The next section will be devoted to models that emerge if not all assumptions leading
to formulation are satisfied. Previously it was shown that if the relation between basic
processes and actions is important, then the essence of the decision problem might be
represented by an elegant mathematical model fitting into the following formulation:
(2) Find the vector x Є D Ϲ Rn , which satisfies y(x) Є O Ϲ Rm and for which c(x) is minimal.
Vectors x Є D, which satisfy y(x) Є O, are called admissible or feasible solutions to the
decision problem. The arguments leading to this formulation include some assumptions that
are clearly not always satisfied. Perhaps the most critical assumptions are that there is only
one criterion and that this criterion can be represented by a real valued function of the
decisions. The first assumption will immediately be weakened by introducing k criteria that
are represented by k real-valued functions C 1(x); C 2 (x); : : : ; Cx (x) on the decisions. Hence
the aim will be to find a feasible x that makes all k criterion functions as low as possible in
some sense. However, we first discuss the more classical case with k = 1.
 Report Model-Based Decision support
10
system

 Single Criteria

If a decision problem has been reduced to the formulation (2), then the only remaining
problem seems to be a matter of mathematical technique, i.e., how to find the minimizing x
among the feasible ones (see e.g., Fletcher, 1987, or Chapter 7 for more details). Actually,
there might still be two difficulties of a conceptual nature. First, the minimizing x might be
nonexistent. And second, there might be more than one minimizing x, leaving open the
selection problem. The first difficulty can be solved by accepting an x that is only
approximately minimal or approximately admissible. The second difficulty may, for instance,
be solved by applying a secondary criterion to the set of minimizing x. This secondary
criterion may be meaningful in the context of the problem or it may just be technical. Hence,
let us assume for the moment that those two difficulties will not bother us; what remains is a
purely mathematical or computational problem. The complexity of this problem highly
depends on the dimensions n, m, on the structure of the sets D, O, and on the structure of the
functions y(x), c(x). In the simplest situation, we have just one decision variable x = x, e.g., a
dimensioning parameter that may be chosen from the positive real numbers, i.e., D = (0;∞),
whereas the consequences of the choice are all represented by the costs, i.e., y(x) and O are
not needed. Let the costs consist of the construction cost, for instance 1 + 2x, and also of
long-range costs of dysfunctioning, for instance, 10e− x. In such a case the solution of (2) is a
simple exercise. More interesting examples emerge from several planning problems with
large numbers of decision variables (n) and large numbers of output functions (m), but
luckily with simple forms for D, O, y(x), and c(x). Typically, D and O consist of intervals for
the individual independent variables xi and dependent variables y j. At the same time, y(x) and
c(x) are often linear in x. In summary, one may conclude that (2) becomes a linear
programming model in this way. For such models, good tools are available even for very large
problems.

 Multiple criteria

Clearly, a weak point in formulation (2) is that it assumes that the quality of decision x ¿ and
x 0can be determined by comparing some real values c(x) and c( x 0). Quite often there are
several such criteria. In this section we will still suppose that each criterion may be
represented by a real-valued function C i(x) on D or on Rn , but extend the scope of the
previous section by accepting that there are k > 1 of such functions. It is unlikely that all such
functions attain their minimum for the same x. Anyhow, if that is the case, then we are
essentially back where we were in the previous section. If that is not the case, then we have
encountered an essential difficulty, which is not just a matter of mathematical technique, but
which requires a new conceptual approach. Let us suppose, for simplicity, that k = 2 and we
have two real-valued criterion or objective functions c ¿(x) and c 0(x), reaching their minimum
at x ¿and x 0, respectively. The added output requirement that y(x) = x 1+ x 2 ≥ 2 and with the
criterion functionsc 0(x) = 2 x 1 + x 2, c ¿(x) = x 2 − x 1represented by dotted lines. In this case we
clearly have x 0 = (0; 2); x ¿ = (5; 1). The basic problem becomes now: shall we simply choose
 Report Model-Based Decision support
11
system

between x 0and x ¿ , or can we construct some compromise solutions between them? In the
latter case, the remaining problem is to find a trade-off between the two criteria. This is not
strictly speaking, a mathematical problem, although it might be made into one with
some extra assumptions on the relative importance of both subproblems.
In this example, one might transform from a criterion into a constraint by adding c ¿ (x) ≤ −1
to the existing constraint y(x) ≥ 2. In this way we get a single criterion optimization problem
with a guaranteed value for c ¿ (x). This would lead to the decision (2,1). It is also possible to
construct a combined criterion out of c 0 andc ¿ . For example, if unit gains in both criteria
would be equally valuable, we could construct a new criterion c(x) by adding the old ones:
c(x) = c 0 (x) + c ¿(x) = (x) = x 1 + 2 x 2:
Minimizing c(x) would give x = (1; 1) straightaway. However, this is based on quite a strong
assumption. Both of the above techniques of dealing with multi-objective optimization
problems (introducing proxy constraints or summing up with importance weights), though
simple and typically suggested by textbooks, have nevertheless some technical and, what is
worse, practical drawbacks, where some other techniques are also presented – more
complicated but giving better practical results. Without introducing new assumptions that
define compromise solutions, we can partition the set of feasible or admissible decisions and
define a subset of such x for which there are surely no better decisions and another subset that
can be improved upon. This can be defined as follows:
Definition 1: A decision x is called a Pareto solution if it is feasible and there is no such ~ x
~
that c i( x) ≤ c i (x) for all i = 1; : : :; k, with strict inequality for at least one i. indicates Pareto
solutions for the simple example presented here with k = 2. Alternative names are efficient or
non dominated solutions, used particularly in cases when we do not minimize (or maximize)
all criterion functions, but minimize some, maximize some and do other things. The set of
Pareto solutions is called the Pareto set; in the criteria or objective space, we call it the Pareto
boundary or Pareto frontier. Clearly, it is reasonable to restrict attention to the Pareto set, but
the problem remains how to select one solution from it (Sawaragi et al., 1985). This selection
is a conceptual problem, although, naturally, there is some relation between the chosen
concept and the complexity of the resulting mathematical problem of finding the best decision
within the chosen concept. The presentation has no pretension of completeness; the aim is
just to indicate which complications may occur and which concepts might be used to
overcome them. Sometimes you can face complex decisions.

MORE COMPLEX DECISIONS

The first assumption in developing formulation (2) was that possible decisions may be
represented by vectors x in n-dimensional Euclidean space. This assumption is clearly quite
strong, although there are many cases where it is reasonable. However, there are also many
cases where decisions require another representation; for instance, if one of the decisions to be
taken is the emission of a particular pollutant at a particular spot. This emission might be
represented by the total emission per year (in tons for instance), but there might also be
 Report Model-Based Decision support
12
system

reasons to consider a much more detailed emission profile represented by the emission speed
at each time instant within the year: x(t); 0 ≤ t ≤ T.
In actual computations, such a representation might be replaced by a finite set of variables
representing some time discretization – such as the emissions for each month in the year, or
the emissions for each week or each day. With this type of discretization, we return to our
original formulation, at the cost of a large number of decision variables. Even when
maintaining the decision as a function of the continuous time indicator t, the idea of
formulation (2) is not really spoiled, only the necessary mathematics are more complicated. In
fact, such problems are the subject of variational calculus and the theory of optimal control.
This was only one way of complicating the representation of decisions. There are also
decisions that are not naturally represented by real numbers. If a family has to choose between
renting, buying, and building a house, where each of the three alternatives can either by
realized in a particular town or a neighboring one, then we have six alternatives. One can
number the alternatives from one to six. Or one can indicate each alternative with a pair (a; b),
where a represents the type of house (rent, existing, new) and b represents the town (one town
or the other). However, it is more difficult to relate such a numerical representation to the
constraints and preferences. Let us suppose that we have financial restrictions on the yearly
costs for housing and on the initial costs. Alternatively, we want to minimize the total daily
travel effort for all family members. In such a case we have in fact six scenarios and for each
scenario we can check whether it fits the financial restrictions or not. From the remaining
scenarios we may choose the one with the lowest travel effort.
In this way we remain completely within the optimization framework, we only miss more
elegant optimization techniques and have to consider all alternatives separately and explicitly.
For this particular case with only six alternatives, such an approach may not be very
cumbersome, but in many cases the number of possible decisions or scenarios may be very
large. A reasonable procedure for such a situation consists in first constructing a restricted
set of representative scenarios and then, after their evaluation, constructing some more
scenarios close to the preferred ones, etc. Also in this way we remain quite well within the
optimization approach. In fact, we develop a kind of search procedure aiming at the optimum.
Another approach would be to apply a divide-and-conquer strategy by first deciding on the
town and only afterwards on the type of house. However, this poses the difficulty of requiring
a comparison of travel efforts without having full details of the location within each town, as
these details are determined by the type of house that will be chosen. One might agree that the
framework of Section 1.3 is most elegant, because within that framework there is a formal
link between decisions and consequences through the functions y(x) and c i(x). However, such
an approach requires that the possible decisions can be represented by some mathematical
entity where functions can be constructed in a natural way. The most suitable candidate for
such an entity is the real valued vector, but some more general entities, like vector valued
functions of a real valued parameter, are also possible. In many cases, regrettably, there is no
natural representation for the decisions that allows the construction of functions y(x) and c i
(x).
 Report Model-Based Decision support
13
system

In fact, this is true in most cases where alternative decisions represent structural choices rather
than choices of dimensioning. In extreme cases, alternative structural choices may even
require completely different techniques for evaluating the respective consequences. As stated
before, this type of difficulty may destroy the elegance of the mathematical programming
approach, but it does not destroy the underlying concept of relating decisions with
consequences. Many decision problems combine both structural and dimensioning decisions.
It seems natural to use a hybrid approach in such cases, where the best dimensioning
decisions are chosen for a structural choice by applying the mathematical programming
approach, whereas the relevant structural choices are evaluated and compared by scenario
analysis. For scenario analysis, many mathematical techniques are available. Usually, the
scenario approach makes it possible to include more detailed simulation models. However,
the evaluation of all possible scenarios is clearly an allusion, whereas in mathematical
programming we can – at least in principle – evaluate and compare all feasible solutions.

UNCERTAIN CONSEQUENCES

Once a decision has been made, all relevant consequences are clear, at least after some
computational processing. But this is surely not always the case. If consequences depend on
the weather (rainfall or amount of sunshine), then those consequences become uncertain in the
sense of depending on chance events. If consequences depend on processes that are not well
understood (like adaptation of animals to changing circumstances), then those consequences
become uncertain in the sense of lack of understanding. This division in two types of
uncertainty is very helpful, but not very strict, as a big grey zone exists between these two
types. In the case of uncertainty depending on chance, it seems natural to extend the
mathematical programming approach with chance elements. This can be done by adding
chance elements to the functions y(x) and c i(x). For instance, in the case of linear
programming, the constraints may be formulated as: Ax ≤ b;
Where A is a n x m matrix, x Є Rn ; b Є Rm . In this case, elements of the matrix A and/or the
vector b may be considered as random variables. One approach would be to require that
inequality (1.1) is satisfied with a certain probability only for the chosen decision x. This is a
relaxation of (1.1). In another approach, a secondary decision vector z is supposed to ensure
that the consequences are not too bad if the realization of A and b appear to be unfavorable.
For the one-criterion case we obtain the following minimization problem:
minimize c T x +d T z (1.2) under constraints: Ax + Bz ≤ b; x ≥ 0; z ≥ 0; where x has to be
chosen before the realizations of A and b are known, but z may be chosen afterwards; hence z
is actually a function of x; A; b. These approaches belong to the field of stochastic
programming (see Ermoliev and Wets, 1988). Naturally, it is much more difficult to include
lack of knowledge. The most straight forward way of including this feature is through the
concept of fuzziness and, indeed, fuzzy linear programming is a well-developed field
(Zimmermann, 1987). Another approach uses scenario analysis and tries to identify the
particular
 Report Model-Based Decision support
14
system

uncertainties related to each scenario. This identification is exploited next to

indicate the possible ranges of consequences.

COMPLICATIONS WITH THE PREFERENTIAL STRUCTURE

It was assumed that the preferential structure could be modeled by putting bounds on some
consequences and designating others as criteria. The strength of this choice is that different
types of consequences can be treated separately. For each type of consequence it may be
decided either to put bounds on it or to consider it as a criterion. In this way each type of
consequence is measured on its own scale, which may be investment costs, levels of pollution,
likelihood of reaching a threshold, etc. The main difficulty remains in finding a decision that
strikes an acceptable balance between the different types of consequences.
The basic assumption is that all types of consequences can somehow be measured on a scale;
however, that is not necessarily true for all types of consequences. If some of the
consequences cannot be measured on a clear scale, one may operate in either of two ways.
First, one may choose not to accept this feature and try to develop an artificial scaling
function. For such a procedure, multi-attribute utility theory provides support. Second, one
may accept this feature and try to live with it. In that case one should accept that some types
of consequences have hard information available while others do not. If one accepts this
feature and acts accordingly, then the feature does not provide real complications. Indeed, it is
possible to integrate hard and soft information on consequences into a well-structured
decision process.

COMPLICATIONS WITH THE DECISION MAKING PROCESS

It might appear, we assume that a single person is responsible for making decisions, whereas
in practice, many decisions are outcomes of complex explicit or implicit decision-making
processes. This assumption arises because the approach of Section 1.3 is frequently presented
as a complete approach, solving the entire decision problem. Such an approach, however, is
not realistic. In fact, the approach can be embedded in a complex decision-making process, as
will be illustrated in various chapters, in particular, the third part dealing with environmental
applications. Of course, such an embedding has consequences for the implementation of the
approach and, therefore, it is of utmost importance that the modeler knows the intricacies of
the decision-making process and understands the role that will be played by the mathematical
programming model. If this condition is not satisfied, the exercise is doomed to be a failure.
Unfortunately, such failures are usually attributed to the approach adopted rather than to the
lack of expertise of the modeler. Therefore, much emphasis will be given to this feature
throughout the book.
 Report Model-Based Decision support
15
system

COMPLICATION WITH MODELING THE BASIC PROCESS

There are cases where the basic processes are already so complicated that one would prefer
not to model them explicitly. In some cases the detailed modeling can be avoided by
modeling on an aggregate level (the treatment of the acid rain problem in Chapter 13 is a good
example of this). In other cases decomposition is helpful, but situations remain where there is
no obvious way out within the mathematical programming approach. Even then, there are still
several alternatives. Here we will mention three of them:

Scenario Analysis
The first alternative is scenario analysis. By formulating some scenarios, one obtains the
possibility to evaluate realistic cases. Even when detailed modeling is too complicated to be
feasible, it is often possible to evaluate particular scenarios by combining partial models and
expert evaluation.

Rule-Based Systems
The second alternative is to use a rule-based system. This implies that the basic processes are
not modeled analytically but evaluated logically by a system that imitates the argumentation
of an expert; thus, a logical type ofmodel is used instead of an analytical type.

Neural Nets
The third alternative is to use a neural net. This means that some black box is trained, with the
use of examples, to give the evaluations. While the latter two alternatives are more suited to
the operational type of decision making, they can also be helpful as building blocks for parts
of decision support systems aiming at more tactical or strategic decisions.

CONCLUSION

The development of last decade showed us understanding of decision support systems is that
complicated decision problems require a form of support that some how integrates various
aspects of decision problems: the relations between basic processes and decisions, the
information, and the decision making process itself. The second conclusion is that most
general approaches to decision support are essentially based on only one of the main aspects.
However, this does not imply that these approaches are useless. It has been stressed already
that it is essential to be aware of the decision-making process when implementing an
approach based on the relations between basic processes and decisions. When doing so, it is
possible to develop models (including optimization models) that can play a useful role in the
decision-making process. The same holds for the relation with information.
 Report Model-Based Decision support
16
system

Many decision support systems are especially made for particular decision situations.
In such systems, one often sees a well-balanced emphasis on all three aspects. One could say
that the main thesis of this book is that it is possible to develop tools and techniques for
constructing well-balanced decision support systems based on substantive models.

REFERENCES
Arnott, D., & Pervan, G. (2006). Eight key issues for the decision support systems discipline.
In Adam, Brezillon,
Carlsson, & Humphreys (Eds.), Creativity and innovation in decision making and decision
support. London:
Decision Support Press.
Morgan, G. (1986). Images of organisations. London: Sage Publications.
Teng, J., & Galletta, D. (1991). MIS research directions: A survey of researchers’ views,
Database, Winter/Spring,
53-62.