Graph based framework for designing Modern Decision Support Systems by editorijettcs


More Info
									   International Journal of Emerging Trends & Technology in Computer Science (IJETTCS)
       Web Site: Email:,
Volume 1, Issue 2, July – August 2012                                          ISSN 2278-6856

     Graph based framework for designing Modern
              Decision Support Systems
                                    Prof (Dr.) M. P. Thapliyal1, Sandeep Kautish2
                                     Associate Professor, H.N.B.Garhwal University Srinagar,
                                         Dist.Pauri Garhwal (Uttarakhand) India 246174
                                                       Research Scholar

                                                                  between model         subsystem   and model deployment
Abstract: As the business world growing at rapid pace,            subsystem [3].
Decision Support Systems are the most talked topic since past
decade. Data contained on Information Systems are used to         In against any user query, search will be made in database
derive analytical models which help decision makers in terms      and will be retrieved if information available. Otherwise,
of simplifying semi-structured problems and suggesting
                                                                  the model database will be searched for seeking models
suitable solutions. The most widely used approach to solve
                                                                  which can produce desired information. If model
semi-structured problems is to combine the expertise of a
decision maker and analytical capabilities of Decision
                                                                  available then there will be another check for its required
Support Systems (DSS), empowered by a model database.             inputs (Liang 1985). If no desired model is available,
This paper aims to frame modern and flexible DSS which            there will be attempt to develop composite model. This
supports knowledge sharing among users and offers improved        implementation will be based on graph theory.
cognitive user profile with desired customization.
Methodology used to design such system is constructing            2. DECISION SUPPORT SYSTEMS (DSS)
customized graph based framework of model section for DSS.
Results suggest that the proposed approach would be viable        The term decision support system (DSS) can be broadly
for use in real world problems.                                   defined as a class of computer based information systems
Keywords: Decision      Support  Systems, Model                   that support decision-making activities. Turban (1995)
Management Systems, Graph Framework, Modeling                     defines it as "an interactive, flexible, and adaptable
                                                                  computer-based information system, especially developed
                                                                  for supporting the solution of a non-structured
1. INTRODUCTION                                                   management problem for improved decision making. It
                                                                  utilizes data, provides an easy-to-use interface, and allows
Researchers are always been trying to find new                    for the decision maker's own insights". DSS are usually
techniques to help decision makers for decision making            context and task-specific and this is the reason why it is
problems of real world scenario. The Decision Support             impossible to give a precise definition including all the
Systems (DSS) is the domain area in which such semi-              facets of the DSS. The only thing that can be said for sure
structured problems are dealt and since evolution of DSS,         about a DSS is that it includes a decision-making process,
it has been emerging topic in computer science study.             which has a more widely accepted definition [5]/
This section describes the concept of Decision Support            Decision-making is the cognitive process with
Systems and intelligent computing techniques i.e. Graph           psychological construct in which users select one course
based approach for model selection and determination.             of action from among multiple alternatives. Only the
The architecture consists of four main components: a              result of the process can be observed in the form of a
database, a model management subsystem, dialogue                  commitment to take action. Decision-making is an
system and knowledge management subsystem. Since the              important part of many professions, where specialists
approach used in the study is based on reasoning and              apply their expertise in a given area to making informed
judgment, a different Model base management system is             decisions.
proposed in the study. The new architecture of model              Decision support systems operate in a multidisciplinary
management system is presented which consists of 3                environment, including among others database research,
modules: a modeling subsystem, a modeling deployment              artificial intelligence, human-computer interaction,
subsystem and an inference engine. The proposed                   simulation methods, and software engineering.
framework includes two major subsystems: modeling and
model deployment system out of which modeling system              Figure 1 shows an overview of a typical DSS, comprising
concentrates on improving effectiveness of models and             five key elements: (i) access to external databases, (ii) a
model deployment subsystem focuses on proper                      Database Manager, (iii) a Model Database Manager, (iv)
utilization of models. Another important component of             a Dialogue Manager and (v) the Decision Maker.
framework is inference engine which makes connectivity

Volume 1, Issue 2 July-August 2012                                                                                  Page 45
   International Journal of Emerging Trends & Technology in Computer Science (IJETTCS)
       Web Site: Email:,
Volume 1, Issue 2, July – August 2012                                          ISSN 2278-6856

                                                            Decision making is the process of getting final state
                                                            (output information) from initial state (input
                                                            information). The models in model base have two core
                                                            elements: nodes and edges. The modeling process based
                                                            on graph is a process which creates a directed graph and
                                                            selects a path which is best or optimum suitable in order
                                                            to for getting desired output. The directed graph or model
                                                            graph consists of all possible alternatives for solving the
                                                            problem by mean of one path for each alternative which is
                                                            called a model.
                                                            Now, various definitions are described for better
                                                            understanding of the terminology used in framework.
                                                              A) Node – A node represents a state or data attributes.

   Figure 1 Overview of a Decision Support System;
       adapted from (Laudon & Laudon, 2004)
                                                            Figure 2 depicts a node named A, which represents a set
                                                                            of data related to supply.
The access to External Databases is used to obtain data
from outside of systems i.e. other information systems.
                                                              B) Edge – An edge represents a set of functionality that
Raw data can be processed to be converted in
                                                              transforms a set of input data to their output. In figure
information, which in turn can be stored in the Internal
                                                              3, E1 represents an edge represents the function of
Database for later use.
                                                              price on demand.
The Database Manager takes responsibility to acquire
                                                            There must be connectivity via an edge between two
external data and mediating the access to Internal
                                                            nodes. In this case only, we can call it a basic model. In
Database. The Model Database Manager stores meta-data
                                                            short, an edge makes a model by connecting two nodes in
about analytical models contained in Model Database and
                                                            particular. Figure 3, as a whole is a basic model as it has
is used to manipulate these models during decision
                                                            2 nodes (A & B) with one edge (E1). We can represent
making processes [4].
                                                            this basic model as a triple (A, E1, B).
The Dialogue Manager provides the user interface layer
                                                              C) Node – An AND node is the ending node of more
that combines the expertise of Decision Makers and
                                                              than one basic model. An AND node is true only if all
system analytical capabilities comprised both in Model
                                                              edges ending on that node are true. Figure 3 represents
and Internal Databases, providing the greatly needed
                                                              a AND Node as node A & B are resulting into node C.
interactivity to DSS.

Research on DSS has increased since past decade but very
few research attempts are made on inclusion of
knowledge sharing concepts. Structured modeling
(Geoffrion 1985) was the theory which focused on
functional relationship among the models during
modeling process [2]. Structure modeling considers each                       Figure 3: AND Node
model of model base as single entity. Geoffrion’s work is
                                                              D) OR Node - An OR node is the ending node of more
the base of graph based approach presented here. The
                                                              than one basic model. An OR node is true only if at
only difference between framework presented here and
                                                              least one edge ending at the node is true. In this
structuring modeling is the latter focuses on functional
                                                              research work, OR node is depicted as ellipse.
relationships among different models and consider high
level of cohesion among these models. The proposed
                                                              E) Path – A path is a finite sequence of edges of the
framework deals not only with representation of model,
                                                              form that –
method of integrating and selecting models.
                                                              a. These edges are connected.
                                                              b. Only one edge is true out of all that enters at each
                                                              OR node.
                                                              c. All edges are true that enters at each AND node.

Volume 1, Issue 2 July-August 2012                                                                             Page 46
   International Journal of Emerging Trends & Technology in Computer Science (IJETTCS)
       Web Site: Email:,
Volume 1, Issue 2, July – August 2012                                          ISSN 2278-6856

                                                               depth-first search, breadth-first search and best-first
                                                               search [9].
                                                               For the purpose of formulation of our model graph, the
                                                               depth-first and best-first search are more suitable as they
                                                               both support modeling concepts. The basic techniques of
                                                               modeling are optimization and satisficing.
                                                               The depth-first search strategy picks an alternative at one
                                                               node arbitrary and work forward. As long as any hope of
                                                               reaching the destination using first alternative, the other
                                                               alternatives remains ignored. In case of the first
                                                               alternative gets proved wrong in desired solution, the
                                                               process starts again with another alternative.
                                                               A typical process of handling user query using depth-
                   Figure 4: OR Node                           search technique is as follows –
                                                                 Step 1: Search for an OUTPUT relation in the model base
F) Composite Model – A composite model is an                     to see whether there is model that produce desired output.
integrated set of set of basic models.                           Step 2: If no such model is found, then stop searching and
                                                                 report about same. The system may ask user to develop new
G)Model graph - A model graph is a graph that                    model.
                                                                 Step 3: If a model is available, search INPUT relation to
represents all possible models i.e. basic model and
                                                                 get input data for execution of the model.
composite models. In a model graph, each path represents         Step 4: Repeat the following sequence until all inputs are
a model. One of the most important properties of model           found or one/more input data is not available –
graph is its acyclic nature.                                     4.1 Take one input and check whether it’s an output of its
                                                                 preceding model.
   3.2 BASIC MODEL                                                         a. If it is true then drop this model and move to
After understanding the graph representation of the              step 3.
model, now, we can portray our basic model. The model                      b. If it is not true then skip this procedure.
is built upon 5 basic types of information: the output of        4.2 Search the database for availability.
the model, inputs required, and computational methods             a. If the input is available in the database, then retrieve its
used in the model, integrity constraints and validity of         value and go to step 4.1
model.                                                            b. If the input not available, then go to step 4.3.
                                                                 4.3 Search the OUTPUT relation of the model to see
All models have 5 basic relations and our model also has
                                                                 whether it can be produced by a model.
same as follows –
                                                                           a. If no model is available, go to step 4.4
                                                                           b. If a model is found, go to step 4.5.
    INPUT (Model#, Inputs)                                       4.4 Prompt the user for the input.
    OUTPUT (Model#, Outputs)                                               a. If it is provided by the user, obtain the value and
    OPERATIONS (Model#, Functions)                               go to step 4.1
    INTEGRITY (Model#, Constraints)                                        b. If not so, then drop the model.
    VALIDITY (Model#, Validation)                                4.5 Search the INPUT relation of the model to find input
                                                                 data required for execution. Repeat step 4 until all input
Here, Inputs and Outputs may have N number of                    data have been obtained or one input in proved
parameters i.e. input1, input2…..input n and so forth.           unavailable.
The context free model presented above can be mounted            Step 5: If all input data for model execution is available,
                                                                 the check integrity constraints.
on the graph based implementation described in previous
                                                                 5.1 If any of integrity constraint is not satisfied, then drop
section. The Inputs and Outputs relations are formulating
                                                                 the model.
a model graph. The Operations relation is represented as         5.2 If all constraints are satisfied, then add the model to
an edge. The integrity relation of the model specifies           the model graph.
constraints which must be satisfied before considering the       Step 6: Check whether any other model producing the
model as applicable. The validity relation is essential for      desired information.
measuring the fitness of the model for a particular type of      6.1 If there is another model then go to step 3.
problem. Validity of a model gives it a subjective               6.2 Otherwise, stop the process.
confidence about the proposed successful implementation
of the model [6].                                                 3.4 MODEL SELECTION TECHNIQUES
                                                               In the previous section, we have devised formulation of
  3.3 MODEL FORMULATION                                        model graph. In this section, we will discuss the strategies
Formulation of model graph requires an extensive search        for providing necessary advice. There are two different
in database and the model base as well. Many approaches        strategies for this: optimizing and satisficing [8].
are developed for creating and traversing a search tree i.e.

Volume 1, Issue 2 July-August 2012                                                                                    Page 47
   International Journal of Emerging Trends & Technology in Computer Science (IJETTCS)
       Web Site: Email:,
Volume 1, Issue 2, July – August 2012                                          ISSN 2278-6856

The optimizing strategy requires a complete model graph        problem. We need one quantitative measure of validity in
and then evaluation will take place to find the best           order to select best model in model graph [6]. Therefore,
alternative. Given the validities of all models are            we need a model evaluation function which will
available; the optimizing strategy will maximize the           determine the validity of model on the basis of some
validity of the selected alternative. Although the             predefined criteria. There are three points need to be
optimization strategy ensures that, given the criteria, the    taken into consideration while developing model
formulated model is the best available alternative,            evaluation function. First is, the criteria for determining
sometimes it becomes unreal to for a system to develop         validity values. We can have four possible criteria as 1)
complete model graph. For example, if a model has b            degree of accuracy of model, 2) user’s preference, 3) costs
branches with d level of depth, depth-first search             incurred, and 4) numbers of models integrated in (Liang
technique will raise O (bd) time complexity to find the        & Jones, 1988).
optimized solution (Korf 1987). Hence, in such kind of         The following is the roadmap in order to reduce
situations, we need to adapt satisficing strategy which        complexity of the model evaluation function.
requires evaluation of each path as soon as it is found and    Step 1: Determine the validity of each model. VALIDITY
accept it if it satisfactory. Hence, formulation of complete   function will validate each model.
model graph may not be mandatory [11]. The satisficing         Step 2: Remove dominated alternatives and simplify the
strategy is not much different from others except it           problem i.e. if more than one model produces desired
evaluates each path at same time of its formulation. The       output then select the one with highest validity.
process of formulating model graph is terminated               Step 3: Calculate validities of all alternative paths from
immediately after finding a satisfactory path. The selected    initial state to the final state.
path will be developed further to produce the desired          Step 4: Select the model with highest validity. In case of a
output. Figure 5 depicts the process of modeling to            complete model graph has been formulated and validities
implement the satisficing strategy.                            of more than one model are equal then we can select
                                                               model on the basis of some other non technical
                                                               constraints i.e. modeling time or machine time. If the
                                                               satisficing strategy has been implemented and complete
                                                               model graph is not available then selection can be made
                                                               on the basis of predetermined cutoff value.

                                                               4. ANALYSIS
                                                               We have devised one framework for creating and utilizing
                                                               Model Base management system. The process starts with
                                                               presenting notions used in the formulation i.e. node,
                                                               graph. After this, the basic model is being defined with
                                                               five kinds of basic operations. Model formulation is the
                                                               next task in the process in which an appropriate model is
                                                               being searched which suits requirements of the particular
                                                               problem. Depth first search algorithm is used to formulate
                                                               the model graph which handles user queries.
                                                               The last section presents the model selection techniques
                                                               for which satisficing technique is used. One model
                                                               evaluation function is being incorporated to validate the
                                                               usefulness of the model in case of multiple models suits to
                                                               one given problem.

                                                               5. CONCLUSION
                                                               This paper has presented an approach to provide DSS
                                                               with flexibility to problem characteristics and adaption to
                                                               user cognitive profile. This approach was comprised by a
      Figure 5: Modeling with satisficing Strategy             method that employs user cognitive profile information
                                                               for decision models creation
One of the major issues involved in implementation of          The current version of the modern DSS employed only
satisficing strategy is the development of model               Decision Trees to solve classification databases. It is
evaluation functions. This need arises when more than          important to highlight that the proposed approach is
one model is suitable for some specific problem. The           abstract and thus independent of technique and class of
Model Management System needs validity values of all           problem. Results shown could be further improved by fine
models to determine which model is best for given              tuning algorithmic parameters Also, other classification

Volume 1, Issue 2 July-August 2012                                                                               Page 48
   International Journal of Emerging Trends & Technology in Computer Science (IJETTCS)
       Web Site: Email:,
Volume 1, Issue 2, July – August 2012                                          ISSN 2278-6856

techniques could be employed to further improve system                of International Joint Conference on Neural
accuracy and capability to deal with different problems.              Networks-IJCNN, 2008, Hong Kong, China.
For example, even though not eloquent regarding                  [13] Quinlan, R. (1993). C4.5: Programs for Machine
explanations about classifications performed by Artificial            Learning, Morgan Kaufmann Publishers, San
Neural Networks (Haykin, 1994) they could be used to                  Mateo, CA.
double check if a classification is correct.                     [14] Zitzler, E.; Laumanns, M & Thiele, L. (2008).
                                                                      SPEA2:      Improving   the   strength  pareto
REFERENCES                                                            evolutionary algorithm, Technical Report 103,
                                                                      Computer Engineering and Networks Laboratory
                                                                      (TIK), Swiss Federal Institute of Technology
  [1] Aitkenhead, M. (2008). A Co-evolving Decision
                                                                      (ETH), 2001.
       Tree Classification Method, Expert Systems with
       Applications, n. 34, 2008, p. 18-25.
  [2] Breiman, L.; Friedman, J.; Olshen, R. & Stone, C.
       (1984). Classification and regression trees,
                                                               Prof. Dr. M. P Thapliyal is M.Sc in Electronics and
       Wadsworth, 1984.
                                                               Ph.D in Physics. He is working as Associate Professor in
  [3] Chacrkaborty,I; Hub, J. & Cuid, D. (2008).
                                                               Computer Science Department of HNB Garhwal
       Examining the effects of cognitive style in
                                                               University, Srinagar Garhwal, Uttrakhand. He has more
       individuals' technology use decision making,
                                                               than 15 years of teaching experience.
       Decision Support Systems, Vol. 45, 2008, pp. 228-
       241. Decision Support Systems, Advances in 290
                                                               Sandeep Kautish has M.Sc in Computer Science, MCA
  [4] Chiavenato, I. (2004). Administração nos novos
                                                               with over 08 years of experience in teaching. He is
       tempos, 2nd Ed., Elsevier, Rio de Janeiro, 2004.
                                                               research scholar for Ph.D. in Computer Scinece.
  [5] Deb, K.; Pratap, A.; Agarwal, S. & Meyarivan, T.
       (2000). A fast and elist multi-objective Genetic
       algorithm: NSGA-II, KanGAL technical report,
       Indian Institute of Technology, Kanpur, India,
  [6] Djamasbi, S. & Stronga, D. (2008). The effect of
       positive mood on intention to use computerized
       decision aids, Information & Management,Vol. 45,
       2008, pp. 43-51.
  [7] Eberhart, R. & Kennedy, J. (1995). A new
       optimizer using particle swarm theory, Proceedings
       of the Sixth International Symposium on
       Micromachine and Human Science, Nagoya,
       Japan. p. 39-43, 1995.
  [8] Laudon, K. & Laudon, J. (2004). Sistemas de
       informação gerenciais: administrando a empresa
       digital, 5ª Ed., Pearson Prentice Hall, 2004.
  [9] Moreau, E. (2006). The impact of intelligent
       decision support systems on intellectual task
       success: An empirical investigation, Decision
       Support Systems, n. 42, p. 593-607, 2006.
  [10] Newman, D.; Hettich, S.; Blake, C. & Merz, C.
       (1998). UCI Repository of machine learning
       databases, University of California, Irvine, Dept. of
       Information and Computer Sciences. USA, 1998.
  [11] Oliveira, F. & Lima Neto, F. (2008). An
       Evolutionary Approach to Provide Flexible
       Decision Dialogues in Intelligent Decision Support
       Systems, Proceedings of the 8th International
       Conference on Hybrid Intelligent Systems (HIS
       2008), Barcelona, Spain, 2008.
  [12] Pacheco, D.; Oliveira, F. & Lima Neto, F. (2008).
       Including Multi-objective Abilities in the Hybrid
       Intelligent Suite for Decision Support, Proceedings

Volume 1, Issue 2 July-August 2012                                                                            Page 49

To top