2. What is a DSS
Use combinations of analytical tools to
assist in decision making.
Tools commonly used include:
Databases
Spreadsheets
Expert Systems
Neural networks
3. Features of a DSS
Handle large amounts of data from
different sources
Flexibility in the reporting of analysis and
the presentation of information / data
The ability to obtain more information
about a solution provided by a DSS (this is
known as ‘drill-down analysis’)
4. Features of a DSS
The ability to perform complex data analysis,
often based on statistical processes
Supports optimization, satisficing and heuristic
solutions to problems
Optimizing – finding the best solution for given criteria
Satisficing – finding a good solution, but not
necessarily the best solution
Heuristic – use of guidelines to find a good solution,
but again, not necessarily the best
5. Features of a DSS
The ability to perform “What if” simulations
and goal seeking analysis
“What if” simulations – making changes to
data and observing the results. This is often
implemented using spreadsheets.
Goal seeking analysis – the process of
determining the initial data required to
produce the desired result
7. Situations
In order to develop a DSS, the nature of the
situation must be determined
Each situation will have peculiarities of data and
require a different data model and a different
type of data analysis
Broadly speaking, there are three types of
situations:
Structured
Semi-structured
Unstructured
8. Structured
Solutions can be clearly identified
The solution process can be automated
All possible inputs are known
All outcomes can be calculated exactly
using an algorithm
All outcomes can be ranked in order of
importance
e.g. Calculating the amount of carpet needed
for a room
9. Semi-Structured
There is usually some method which can be
used to reach a solution
However, the solution is not guaranteed
The data may not be clear or complete
Occasionally, experts examining a semi-
structured situation will reach different
conclusions
Most real world problems are semi-structured
e.g. Betting on the horses.
10. Unstructured
Very difficult to describe or understand the
problem
Very difficult to identify and evaluate alternative
solutions and select a solution
Often, no clear method for solving the problem
Judgement and experience of human experts
are often used
e.g. Predicting the state of the world’s economy in three
years time
11. Examples of DSS
All of the textbooks I have examined do a
particularly good job on this part of the
syllabus
When considering the various types of
DSS there is scope to do some practical
work
Obviously the class can cover databases
(again!)
12. Examples of DSS
Other kinds of practical work include:
Spreadsheets – a really useful resource for
examining structured and semi-structured
situations
Expert Systems – There are a few options
here
Neural Networks – Beyond doing internet
research, there is not many practical tasks
students can do
13. Organising and DSS
Spreadsheets – Most spreadsheets will
have the following organisation
Title Area – inc purpose, author, date
Data Labels
D F
Data
a o
t r
a m
L u
a l
b a
e e
l
s Formulae
14. Spreadsheets
All of the material on spreadsheets from the Year 11
course should be revised
Students should also cover new material, such as
templates and multiple worksheets
The real challenge is to get students to analyse a
problem and then design the spreadsheet solution
This is a real test of their understanding. Many students
can create a spreadsheet by following a series of steps.
They may even understand what is going on. However,
to create a spreadsheet from scratch requires a depth of
understanding and a lot of practice.
15. Spreadsheets
For an able class it is also worthwhile teaching
goal seeking and formula auditing in Excel
Graphing data is another area where students
need a lot of practice
Graphs are an important component of DSS as
they enable trends/patterns to be identified
Students must remember to include a title for a
graph, label the axes, use an appropriate scale
and choose the appropriate type of graph to suit
the data
17. Expert Systems
Used to provide
Expert quality advice
Diagnosis
Recommendations
Preserve an experts knowledge
Ultimately, a human being is required to
judge any solution made by an expert
system and make any final decisions
18. Why use an Expert System?
Experts may be in short supply
Advice from experts may be expensive
It may be difficult to access an expert
Example. A typical GP has only
rudimentary knowledge of tropical
diseases. To get a proper diagnosis, you
would have to consult with a specialist or
an expert system
19. How do we build an expert system?
A Knowledge Engineer needs to consult
with human experts to determine the
heuristics (rules of thumb) needed for the
system
Often these rules are not certain and are
expressed using ‘fuzzy logic’
20. Structure of an Expert System
Typically an expert system has three parts:
A database of facts – about the current problem and
the current status of the problem-solving process
A knowledge base – this contains condition/action
rules (IF/THEN)
An inference engine – applies the facts and rules to
determine a solution. Inference engines use two
methods to reach a solution…
Forward Chaining
Backward Chaining
21. Forward Chaining
The user supplies all the data
The inference engine uses the supplied data
and rules to work forward to see what
conclusions can be reached
This is often used where there is no clear goal
and the system is attempting to determine all
possible conclusions implied by the data
XCON is an expert system used to configure
computers that uses forward chaining
22. Backward Chaining
The inference engine starts with the
conclusion
It looks for rules/ facts that supports the
goal
If all the rules and facts are found to be
true, then the conclusion/goal is correct
MYCIN – one of the first medical expert
systems used backward chaining
23. Artificial Neural Networks
An ANN is a system modelled on the
human brain
They learn from experience and are built
in such a way that often the designers are
unable to explain the conclusion/decision
that has been reached
Used in artificial vision, voice recognition
and DSS
24. Artificial Neural Networks
An ANN can
be taught to recognise patterns
adapt to changing environments
generalise
make decisions based on prior experience
get better at a task as they gain more experience
Data supplied to an ANN is usually not exact
and is given weightings to determine the relative
importance of the data to the total input
Ideal for unstructured situations, however they
are slower at obtaining a solution than
conventional computing systems
25. Structure of an ANN
ANN’s make use of parallel processing:
A problem is broken down into a number of
smaller components
Each sub-problem is processed
simultaneously by a number of special CPU’s
called PE’s or Processing Elements
A PE is a lot simpler than a CPU and can only
perform a limited range of tasks. However…
26. Structure of an ANN
There are millions of PE’s in an ANN
The PE’s are interconnected in a network
structure so that a PE can only pass data to
those PE’s to which it is connected
The overall goal is for the network to replicate
the function of the human brain, with each PE
operating like a nerve cell
The connections between PE’s can be altered
by the system to match the nature of the
problem. Thus learning occurs
27. Structure of an ANN
Initially, ANN’s have to be taught how to solve a
problem
As experience grows they are able to solve more
problems independently
This technology is still in its infancy
Since the complete functioning of the human
brain is yet to be completely determined (if ever),
it remains to be seen as to whether a truly
intelligent machine will ever be developed
28. Fuzzy Logic
When developing an expert system one of the
most important steps involves developing the
knowledge base of rules
These are usually in the form of
IF (condition) THEN (action)
However, often the rules we develop are not
clear cut
We have to associate a probability with the rule
because sometimes a piece of information is
best described as fairly true or fairly false
29. Fuzzy Logic
The development of these kinds of rules is what
fuzzy logic is all about
The probabilities that we attach to the rules are
called certainty factors (CF) and are based on
historical data and statistical analysis
e.g. IF (Stella is tired) THEN (She will be in a bad mood)
(CF=0.7)
Of course any conclusion reached by the expert
system will also have a probability associated
with it
30. Social and Ethical Issues
While expert systems can be claimed to
preserve an expert’s knowledge, it can be
argued that they counter the necessity of
having a human expert, i.e. they “dumb
down” the job
Data Mining
Erroneous inferences
Privacy
Responsibility for decisions
32. Practical Activities
One textbook suggests that students download a
DSS demonstration software package and do an
evaluation based on test data
I have tried to find such demonstration packages
with limited success
On the one hand there are expensive systems
available, but no demo. On the other hand there
are simple systems, like a loan repayment
calculator that are too trivial to warrant serious
consideration
Any suggestions in this regard would be
appreciated
33. Practical Activities
A major focus could be on using
spreadsheets
There are some really interesting
simulations you can do with Excel. A
good source of activities is the book
“Mathematics at Work: Modelling your
World”
34. Practical Activities
Similarly, a few textbooks have suggested searching the
web for demo expert systems
Most of the downloads I have found are of poor quality
and of little interest to students
CLIPS – is not a bad expert system shell. However, it
does take a while to fully understand its operation
Freeware versions of Prolog can also be used to develop
a simple expert system