Cost Estimation Using AI Techniques

1
Cost Estimation Using AI Techniques
Arab Academy For Science, Technology Maritime
Transport
College Of Computing Information Technology
Supervisor Prof. Dr. Meer Hamza Speaker
Ahmed S. ALSUM Teaching Assistant in FCIS,
Mansoura University
MASTER
2004
2
OutLine

• Cost Estimation Techniques Overview.
• Software Cost Estimation with Case Based
  Reasoning.
• Software Cost Estimation with Neural Networks.
• Software Cost Estimation with Fuzzy Models.

3
Importance of Cost Estimation!

• Budgeting
• Tradeoff and risk analysis.
• Project planning and control.
• Software improvement investment analysis.

4
Generic Vs. Bespoke

• With products developed for the mass markets the
  cost of development is not visible in the price
  of the product.
• With custom or bespoke development, the
  development cost influences the price.
• It is of strategic importance for an
  organisation, whether as a customer or a
  developer

5
From surveys of organizations

• only 25 of projects come within their originally
  predicted cost and schedule,
• But 66 of companies significantly underestimate
  the time and cost and that the costs for similar
  projects can vary by up to 200.

6
Software Estimation Techniques
Expert judgement is still the most dominant
method of estimation. (1991)
7
Why we use AI techniques in Cost Estimation?

• Software project cost estimation is difficult
  because of problems of quantifying project size
  and because of the continual emergence of new
  technology.

8
Software Cost Estimation with Case Based
Reasoning CBR
9
What is CBR?

• Case-based reasoning (CBR) may function as a
  corporate memory where experience is stored as
  cases and may be drawn upon when new problems
  arise.

10
Should we use CBR?

• Software project cost estimation is difficult to
  derive accurate size and cost figures from the
  features of a project that are known early.
• This is characteristic of weak theory domains
  where CBR has potential as a solution.

11
CBR Techniques?

• CBR involves matching the current problem against
  similar problems that have occurred in the past.
• It uses past experiences to solve new problems
  which corresponds to how experts operate.

12
How To Apply CBR?

• The current research focuses on problem domains
  that are
• similar in nature,
• similar types of applications
• from similar organizations

13
How to say its similar?

• Examination of the features or attributes that
  contribute to the case representation used in the
  current research imply that
• Using attributes such as numbers of inputs,
  outputs, internal and external files to a
  identify similar cases.

14
Early Estimation
We are here
15
Early Estimation Problem

• The main problem is the availability of enough
  predictive features that can give an accurate
  estimate of effort.
• To be SUCCESSFUL in applying CBR to early cost
  estimation will involve identifying more abstract
  features than metrics such as size or specific
  detailed design features to represent a case.

16
CBR in early estimation How to find a measure of
the system size.

• Effort 8 Size Productivity
• Solution describe using analogy for effort
  estimation and state that it is important to
  choose at least one variable or feature to act as
  a size driver.

17
Cost Driver

• Cost drivers are the variables that are believed
  to influence the cost of software development.
• Ex.. COCOMO uses 15 cost drivers
• Problem with cost drivers are
• There is a lack of clear definitions for them
  especially quality, complexity and experience,
• Difficult to quantify and can be very subjective.

18
Case Representation
19
It is only a management representation?

• Management can have more influence on the
  productivity of the programming staff than any
  technology now in use
• Research has also identified the importance of
  significant user participation and experience in
  successful projects

20
You speak about the same domain What about
Estimation Across Broader Domains?

• In order to use CBR in estimation in a broader
  context there is need to identify abstract
  features that will support reminding across
  different contexts and a need to develop
  adaptation techniques to transform cases between
  these contexts.

21
Software Cost Estimation with Neural Network
22
What is Neural Network?

• These are estimation models that can be trained
  using historical data to produce ever better
  results by automatically adjusting their
  algorithmic parameter values to reduce the delta
  between known actuals and model predictions.

23
Ex..

24
To create NN

• developing an appropriate layout of neurons, or
  connections between network nodes.
• This includes defining
• the number of layers of neurons,
• the number of neurons within each layer,
• and the manner in which they are all linked.
• The weighted estimating functions between the
  nodes, ...etc

25
To create NN

• the model must be trained by providing it with a
  set of historical project data inputs and the
  corresponding known actual values for project
  schedule and/or cost.

26
Problems

• Extremely large data sets are needed to
  accurately train neural networks.
• Neural Networks provide little intuitive support
  for understanding the sensitivity relationships
  between cost driver parameters and model results.

27
Conclusion for NN CBR

• Experience to date indicates that neural-net and
  dynamics-based techniques are less mature than
  the other classes of techniques, but that all
  classes of techniques are challenged by the rapid
  pace of change in software technology.

28
Software Cost Estimation with Fuzzy Model
29
Why Fuzzy Model?

• Estimation of effort/cost required for
  development of software products is inherently
  associated with uncertainty. ,
• we are concerned with a fuzzy set-based
  generalization of the COCOMO model, that what we
  called F-COCOMO.

30
Difference BetweenCOCOMO model (f-COCOMO)

• standard COCOMO model the inputs of the include
  an estimation of project size and an evaluation
  of other parameters.
• Rather than using a single number, the software
  size can be regarded as a fuzzy set (fuzzy
  number) yielding the cost estimate also in form
  of a fuzzy set.

31
Fuzzy Sets

• Using fuzzy sets, size of a software project can
  be specified by distribution of its possible
  values.

32
Ex..

• a small software project can be described by a
  fuzzy set K.

In general, a fuzzy set K is described by its
membership function K(x) which expresses the
degree of membership of x to the fuzzy set K
describing a certain concept
33
Basic COCOMO

• Models may have the general form
• where S is the code-size, and a, b are (usually
  simple) functions of other cost factors.

34
A. Simple f-COCOMO Model Triangular Numbers
35
Triangular Numbers
As the input variable (size K) is a fuzzy set .
(fuzzy number), so is the effort E.
where K(x) and E(e) denote membership functions
of the size of the code and membership function
of the effort
36
Triangular Numbers
This formula can be used directly for estimation
of effort from size of software project described
by triangular fuzzy set.
37
Simple f-COCOMO Model
38
Conclusion On Fuzzy Model

• The model admits inputs (say, size of code) that
  are nonnumeric and therefore more acceptable to
  the designer and project manager.
• Fuzzy sets create a more flexible, highly
  versatile development environment.
• First, they help articulate the estimates and
  their essence (e.g., by exploiting fuzzy numbers
  described by asymmetric membership functions).
• Second, they generate a feedback as to the
  resulting uncertainty (granularity) of the results

39
References.

• Sarah Jane Delany, Pádraig Cunningham and
  Wolfgang Wilke, The Limits of CBR in Software
  Project Estimation, 1998 German Workshop on
  Case-Based Reasoning.
• Barry Boehm, Chris Abts, Sunita Chulani ,
  Software Development Cost Estimation Approaches
  A Survey, 1998.
• Petr Musflek, Witold Pcd.rycz, Giancarlo Succi,
  Marek Reformat, Software Cost Estimation with
  Fuzzy Models

40
Thanks
CS MASTER
2004
S