385
ISPRS, Vol.34, Part 2W2, “Dynamic and Multi-Dimensional GIS", Bangkok, May 23-25, 2001
happen if a network of retail outlets is changed, and determining
optimal business strategies in a region.
The incorporation of decision-making models within GIS to
develop a SDSS is a powerful approach for solving spatial
business problems. Business decisions take place in a specific
context, hence generic SDSS are only of limited value. The
analytical power required can only be obtained from a
customised SDSS developed using a set of software modules
and analytical tools that are pertinent in the context of a specific
business application, here the closure of branch banks.
2.1 The need and benefit of Integration
The need for integration in this research is driven by the need to
choose branches for closure, considering research to increase
knowledge can improve the act of making choices and produce
good decisions. Thus the ultimate goal of integration is not only
to develop a better research tool, in the form of more powerful
models and analysis software, but also to aid bank branch
planning and the decision making process.
Parks (1993) argued that integration is not a new idea but rather
the further co-adaptation of existing tools and methods. The
integration might cross-fertilize and mutually reinforce each
other to develop new ways in which they coula be designed to
serve additional users-including those who must participate in
common decision-making process. Three primary reasons for
integration suitable to begin such discussion are given as
follows:
Firstly, 80% of data in the banking industry are spatial
referenced (King, 1993), spatial representation is critical to
complex multiple location choice problems. Decision support in
branch banking location problems requires facilities for the input,
management and output (display) of spatial-referenced data,
facilities that are currently available in existing GIS packages.
But GIS currently lack the predictive and related analytic
capabilities necessary to examine complex problems. The GUI
and spatial operators in GIS have not been developed for
multiple location problems and GIS have not generally been
amenable to decision making tasks without considerable effort to
customise them.
Secondly, the decision making process can benefit from the use
of multi-criteria decision making (MCDM) techniques, which
provide both a sound methodology and platform for decision
analysis and an operational framework for actual decision
making (Roy, 1996). MCDM techniques can be used to rank
branches in terms of whether they are more or less preferable
for closure according to a variety of social, economic, and
demographic criteria. This approach facilitates the decision
making process by making it more explicit, rational, and efficient
(Hobbs et a!., 1992). Decision making tools typically lack
sufficiently flexible GIS-like spatial analytic components and are
often inaccessible to potential users less expert than their
makers. What GIS could offer to decision making is a flexible
environment with a powerful spatial analysis tools, such as
buffer and overlay, and the strong visualisation capability.
Thirdly, GIS and MCDM technology can both be made more
robust by their linkage. The effort to combine the strengths of
these tools will be mutually beneficial. So writing separate
pieces of software for location problems is not a good strategy,
and the optimal option is to take advantage of these facilities
through integration (Goodchild etai, 1992).
2.2 Integration methods
Many ways of integrating external software packages with GIS
have been used in previous research, the architectures of which
relate to the degree of ‘closeness’ involved (see Goodchild et al.,
1992; Fedra, 1993; Goodchild et al., 1993). The generally
accepted classification identifies three levels of coupling or
integration: loose coupling, close coupling, and full integration.
Following the arguments by Jankowski (1995), the loose
coupling architecture is used in this research for linking together
the two software packages. The SDSS proposed in this study is
designed around the integration of the Criterium DecisionPlus
3.0.3 (CDP) software, a commercial decision making support
package from InfoHarvest, with ArcView GIS 3.2 using Dynamic
Data Exchange (DDE) in the Microsoft Windows environment.
Dynamic Data Exchange enables the continuous and automatic
exchange of data between ArcView and CDP by means of a bi
directional data transfer with ArcView as the database and
visualisation engine and CDP as the engine for data analysis
(Fig. 1).
Transfer
Fig.1 GIS-MCDM System Integration
3. METHODOLOGY
Branch closure decision making usually involves many decision
makers, most of whom are likely to have a limited technical
competence and hence little understanding of the complexities
of decision-making models or the way they work. There are
advantages therefore if the formal models used can be quickly
and easily understood, at least intuitively, by those involved in
the decision-making process and it is best if the application of
the models is not time consuming.
A number of different models have been proposed to structure
and solve multi-criteria decision problems and computational
methods developed for their application. In this research, the
analysis is based on the Simple Multi-Attribute Rating Technique
(SMART) model, which is convincible to decision makers
through its rich applications (Corner and Kirkwood 1991).
3.1. The SMART model
SMART was put forward by Edward’s’ in 1971 (Edwards, 1971),
and it is closely related to the multi-attribute utility approach that
had been developed by Keeney (1969). The basic equation in
SMART, which is the formula for a weighted average is:
V t - YjWj u ij (Eq-1)
j
subject to Yj W j = 1
j
where U, is the aggregate utility for the /th alternative
Wj is the normalized importance weight of the /th
attribute of value
u,j is the normalized value of the /th alternative on the
/th attribute
In SMART, the lowest level criteria are called attributes. The
numerical values (called ratings) assigned to these attributes are
derived from value functions. The structure used to model the
