ANALOGY
The analogy cost estimation technique involves comparing
one or more completed projects in a similar domain as a means
of producing new estimates. This can be achieved by analysing
the data collected from these completed projects against the
new project, assessing similarities. As the development
effort is known this is used as a basis for estimating the
current project. Estimation by analogy seems straightforward,
but there are a number of problems that need to be addressed,
how to determine project features ? These are restricted to
information that is available at the start of the estimation
process, examples of these can include application domain,
software process and so forth. Other common problems are how
many projects should we measure against ? Too few projects
can lead to maverick projects being used; too many projects
can weaken the effect of the closest analogies. With this
estimation method the need for an expert is eliminated,
therefore the sometimes difficult knowledge elicitation stage
disappears as analogy systems examine problems that have
occurred, unlike other cost estimation techniques which need
to tackle all possible problems anticipated on a project.
The process involved to predict software cost estimates
can be broken down into the following steps according to [COWDEROY & JENKINS]
1 Selection of
analogies
2
Assessing similarities and differences
3 Assessment
of analogy quality
4
Consideration of any special cases
5 Creating the estimate
The first stage in the process is to actually
compile a database of suitable completed projects that we can
measure similarities against in our new project. The
selection process therefore involves selecting analogous
projects that reflect both the development environment and
project features as the new project.
Now that we have a suitable database of analogous
projects, the next stage is to assess the similarities to the
new project. Similarity is determined in terms of project
features, the number of features depend upon the data
available to characterise the projects. [SHEPPERD et al.] reports
datasets with as few as one project feature and as many as 29
features being used.
The process of selecting analogies can be achieved by the
similarity in n-dimensional space where each dimension
corresponds to a project feature. The most similar projects
will be those closest together in n-dimensional space. The
idea of n-dimensional space can be depicted below, the
diagram has only 3 dimensions as it is hard to express more
than 3 dimensions in a diagram.
Figure 6 N-dimensional space
Figure 6 gives an example of how the selection process
occurs. Lets suppose that we want to measure the effort for
the new project, by analysing project features experience,
function points and effort. From visualising 3-dimensional
space you can clearly see that project B is closest to the
new project and project A would be rejected is this case
because of the substantial distance in dimensional space. The
example here is rather trivial but in essence it is what
happens. A realistic approach would involve more projects and
dimensions or project features being used. This allows you to
select the closest analogies to the new project, ultimately
leading to a more accurate estimate.
To access how well the quality of analogy predications
perform you can apply estimation predication quality
measures:
- absolute error
- percentage error and mean percentage error
- magnitude of relative error (MRE) & mean
magnitude of relative error (MMRE)
- pred(25)
The results from the quality measures can be fed back into
the analogy process to examine the results of why some
estimates are better than others.
Sometimes it may be necessary to exclude some projects
from consideration. Although the project could have
been selected in stage 2 for similarities, that project might
have a project feature that you wish to ignore. For example,
the project under consideration may have used an unusual
design methodology.
Assessment of Analogy
Other Reasons to use Analogy
The whole process of storing analogous projects
eliminating redundant features to finding analogies can be
automated by a PC based software tool known as ANGEL (ANaloGy softwarE
tooL) developed by Chris Schofield and is available as a
shareware copy. This machine learning tool is the focus of a
research project being undertaken at Bournemouth University.
For further information on the this tool, please click on the
angel link.
Further Information
A paper by [SHEPPERD,
SCHOFIELD, KITCHENHAM], discusses effort estimation using
analogy in detail, and is available by clinking on the link
This page was last updated 12/3/97 Dan. Snell,
Bournemouth University, Copyright. 1997