A Fuzzy Group Prioritization Method for Deriving Weights and its Software Implementation

— Several Multi-Criteria Decision Making (MCDM) methods involve pairwise comparisons to obtain the preferences of decision makers (DMs). This paper proposes a fuzzy group prioritization method for deriving group priorities/weights from fuzzy pairwise comparison matrices. The proposed method extends the Fuzzy Preferences Programming Method (FPP) by considering the different importance weights of multiple DMs . The elements of the group pairwise comparison matrices are presented as fuzzy numbers rather than exact numerical values, in order to model the uncertainty and imprecision in the DMs’ judgments. Unlike the known fuzzy prioritization techniques, the proposed method is able to derive crisp weights from incomplete and fuzzy set of comparison judgments and does not require additional aggregation procedures. A prototype of a decision tool is developed to assist DMs to implement the proposed method for solving fuzzy group prioritization problems in MATLAB. Detailed numerical examples are used to illustrate the proposed approach.


I. INTRODUCTION
HERE are various techniques for deriving priorities/weights for decision elements (e.g.attributes/criteria) from a decision maker (DM) or group of DMs, some of which are reviewed by Choo and Wedley [1] and Ittersum et al. [2].Most techniques are based on either direct weighting or on pairwise comparison.In direct weighting, the DM is directly asked to give values between 0 and 1 to each decision element to assign their importance.Some methods for deriving attributes/criteria weights by direct assigning techniques are: the Simple Multi-Attribute Rating Technique (SMART) [3], SWING weighting methods [4], and SMART Exploiting Ranks (SMARTER) [5].
When the DM or the group of DMs are unable to directly assign decision elements' weights, the Pairwise Comparison (PC) method proposed in [6] can be used.
Psychological experiments have shown that weight derivation from PC is much more accurate than direct weighting [7].Therefore, the PC methods are often used as an intermediate step in many MCDM methods, as Analytic Hierarchy Process (AHP) [7], Analytic Network Process (ANP) [8], PROMETHEE [9], and Evidential Reasoning (ER) [10].
The PC methods require construction of Pairwise Comparisons Judgment Matrices (PCJMs).In order to construct a PCJM, the DM is asked to compare pairwisely any two decision elements and provide a numerical/linguistic judgment for their relative importance.Thus, the DM gives a set of ratio judgments to indicate the strength of his/her preferences, which are structured in a reciprocal PCJM.Then, the weights or priority vectors of the decision elements can be derived from the PCJM by applying some prioritization methods.
There are numerous Pairwise Comparisons Prioritization Methods (PCPMs), such as the Eigenvector Method [7], the Direct Least Squares Method [11], the rank-ordering method [7], the Logarithmic Least Square Method [12], and the Fuzzy Programming Method [13].Choo and Wedley [1] summarised and analysed 18 PCPMs for deriving a priority vector from PCJMs.They discussed that no method performs best in all situations and no method dominates the other methods.
However, in many practical cases, in the process of prioritization the DMs are unable to provide crisp values for comparison ratios.A natural way to deal with the uncertainty and imprecision in the DMs' judgments is to apply the fuzzy set theory [14] and to represent the uncertain DMs' judgments as fuzzy numbers.Thus, Fuzzy PCJMs can be constructed and used to derive the priority vectors by applying some Fuzzy PCPMs.Such methods are proposed by Laarhoven and Pedrycz's [15], Buckley [14], Chang [16] and Mikhailov [17], and applied for group decision making.
The existing fuzzy PCPMs have some drawbacks.They require an additional defuzzification procedure to convert fuzzy weights into crisp (non-fuzzy) weights.However, different defuzzification procedures will often give different solutions [17].
The linear and non-linear versions of the Fuzzy Preference Programming (FPP) method [17] do not require such defuzzification procedures, but their modifications for group decision making situations assume that all the DMs have the same weight of importance.However, in real group decision making problems, sometimes some experts are more experienced than others [18][19].Therefore, the final results should be influenced by the degree of importance of each DM.
In order to overcome some of the limitations of the group FPP method, a new group version of the FPP method is proposed by introducing importance weights of DMs in order to derive weights for decision elements in group decision problems.The proposed method has some attractive features.It does not require any aggregation procedures.It does not require a defuzzification procedure.It derives crisp priorities/weights from an incomplete set of fuzzy judgments and incomplete fuzzy PCJMs.Moreover, the proposed method considers the DMs weights.
For applying the proposed method and solving prioritisation problems, a Non-Linear FPP Solver is developed based on the Optimization Toolbox of MATLAB, in order to overcome the complexity of programming.This decision tool is demonstrated by solving a few numerical examples.
The remainder of this paper is organised as follows.In Section II, representation of the fuzzy group prioritization problem is briefly explained.Then, the proposed method is presented in Section III and illustrated by numerical examples in section IV.The developed Non-Linear FPP Solver is presented in section V, followed by conclusions.
, which represents the relative importance of the n elements.

III. GROUP FUZZY PREFERENCE PROGRAMMING METHOD
The non-linear FPP method [17] derives a priority vector


, which satisfies: where  ~ denotes 'fuzzy less or equal to'.If M is the overall number of fuzzy group comparison judgments, then M 2 fuzzy constraints of the type (3) are obtained.
For each fuzzy judgment, a membership function, which represents the DMs' satisfaction with different crisp solution ratios, is introduced: The solution to the prioritization problem by the FPP method is based on two assumptions.The first, requires the existence of a non-empty fuzzy feasible area P ~ on the The fuzzy feasible area P ~ is defined as an intersection of the membership functions (4).The membership function of the fuzzy feasible area P ~ is given by: The second assumption identifies a selection rule, which determines a priority vector, having the highest degree of membership in the aggregated membership function (6).Thus, there is a maximizing solution * w (a crisp priority vector) that has a maximum degree of membership * A new decision variable  is introduced which measures the maximum degree of membership in the fuzzy feasible area P ~.Then, the optimization problem ( 7) is represented as The above max-min optimization problem ( 8) is transformed into the following non-linear optimization problem: The non-linear FPP method can be extended for solving group prioritization problems.Mikhailov et al. [20] proposed a Weighted FPP method to the fuzzy group prioritization problem by introducing the importance weights of DMs.However, the Weighted FPP method requires an additional aggregation technique to obtain the priority vector at different  -thresholds.Consequently, this process is time consuming, due to several computation steps needed for applying the  - threshold concept.Therefore, this paper modified the nonlinear FPP method [17], which can derive crisp weights without using  -threshold and by introducing the DMs' importance weights.
When we have a group of K DMs, the problem is to derive a crisp priority vector, such that priority ratios The ratios j i w w can also express the satisfaction of the DMs, as the ratios explain how similar the crisp solutions are close to the initial judgments from the DMs.The inequality (10) can be represented as two single-side fuzzy constraints of the type (3): The degree of the DMs' satisfaction can be measured by a membership function with respect to the unknown ratio We can define K fuzzy feasible areas, k P ~, as an intersection of the membership functions (12) corresponding to the k -th DMs' fuzzy judgments and define the group fuzzy feasible area

By introducing a new decision variable k
 , which measures the maximum degree of membership of a given priority vector in the fuzzy feasible area k P ~, we can formulate a max-min optimisation problem of the type (8), which can be represented into: For introducing the DMs' importance weights, let us define . For aggregating all individual models of type ( 13) into a single group model, a weighted additive goal-programming (WAGP) model [21] is applied.
The WAGP model transforms the multi-objective decision making problem to a single objective problem.Therefore, it can be used to combine all individual models (13) into a new single model by taking into account the DMs' importance weights.
The WAGP model considers the different importance weights of goals and constraints and is formulated as: In order to derive a group model, where the DMs have different importance weights, we exploit the similarity between the models (13) and (15).However, the non-linear FPP model (13) Where the decision variable k  measures the degree of the


; k I denotes the importance weight of the k -th DM, .,... 2 , 1 K k  In (16), the value of Z can be considered as a consistency index, as it measures the overall consistency of the initial set of fuzzy judgments.When the set of fuzzy judgments is consistent, the optimal value of Z is greater or equal to one.For the inconsistent fuzzy judgments, the maximum value of Z takes a value less than one.
For solving the non-linear optimization problem (16), an appropriate numerical method should be employed.In this paper, the solution is obtained by using MATLAB Optimization Toolbox and a Non-linear FPP solver is developed to solve the prioritization problem.

IV. ILLUSTRATIVE EXAMPLES
The first example illustrates the solution to the fuzzy group prioritization problem for obtaining a priority vector and a final group ranking.The second example demonstrates how the importance weights of DMs influence the final group ranking.

A. Example 1
This example is given to illustrate the proposed method and also the solution by using the Non-linear FPP Solver.
We consider the example in [20], where three DMs ( Using the above data and the non-linear model ( 16), the following formulation is obtained:  This solution can be compared with the crisp results from the example in [20] as shown in Table I.We may observe that we have the same final ranking , from applying the two different prioritization methods.However, the Weighted FPP method [20] applies an aggregation procedure for obtaining the crisp vector from different values of priorities at different  -threshold.While, the proposed non-linear group FPP method does not require an additional aggregation procedure.
If ), as seen in Fig. 2. We can conclude that the average computation time (Minutes) for the Weighted FPP method highly increases as the number of decision elements n increases, compared with the proposed method.Hence, these results showed that the method proposed in this paper is more efficient, with respect to the computation time.Therefore, the proposed method in this paper demands less computation time than the Weighted FPP method [20].

B. Example 2
This example shows that the importance weights of the DMs influence the final group ranking.

  I I
For both situations, the final rankings for both individual DMs are shown in Tables II and III respectively.The final group rankings are also shown in Tables II and III (the third row of each table).The results are obtained by using the Non-Linear FFP Solver.Each final group ranking is obtained by solving a non-linear program of type (15), which includes eight non-linear inequality constraints corresponding to the given DMs' fuzzy comparison judgements.
It can be observed from Tables II and III that the final group ranking tends to be the individual ranking of the a The method proposed in [16] with applying  -threshold.b The method proposed in this paper without applying  -threshold.DM who has the highest importance weights.In more detail, it can be seen from Table II that the judgements of the second DM with the highest importance weight ( 8 .0 2  I ) influence, more strongly, the final group ranking.On the other hand, the final group ranking in Table III is dependent on the first DM, who has the highest importance weight ( 8 .0 1  I ).From examples 1 and 2, we can observe the importance of introducing importance weights of the DMs to the fuzzy group prioritisation problem.It is seen that the final group ranking depends on the DMs' importance weights.

V. SOFTWARE IMPLEMENTATION USING MATLAB
MATLAB is a numerical computing environment, which allows matrix manipulations, plotting of functions and data, implementation of algorithms, creation of user interfaces, and interfacing with programs written in other languages, including C, C++, Java, etc. [22].This development environment includes many functions for statistics, optimization, and numeric data integration and filtering [23].
In this paper, we use the Optimization Toolbox and the Graphical User Interface (GUI) of MATLAB as the development tools for implementing the proposed group nonlinear FPP method, because these tools provide powerful numerical functions, optimisation procedures, good visualisation capabilities and programming interfaces.
Essentially, there are three steps for programming and developing the Non-Linear FFP solver: Step 1: Coding the model into the system.A number of functions are available in the Optimization Toolbox-MATLAB to solve the non-linear programming problem.In our prototype, the optimisation problem is solved using the sequential quadratic programming procedure [19].
Step 2: Creating a basic user interface.In this step, the interface is designed, so that it can run in the MATLAB command window.The aim of this user interface is to obtain the input information from the DMs.
Step 3: Developing the system based on the GUI functions.In this step, the MATLAB GUI functions are employed to develop a more user-friendly system.
Regarding the given data in example 1, the input information which should be acquired includes the total number of decision elements, the names of these elements and the total number of DMs, as shown in Fig. 3.Then, the pairwise judgments for each DM can be entered by the user, as illustrated in Fig. 4. According to example 1, the fuzzy judgments for the DM 1 are illustrated in Fig. 4. Thus, the main feature in the developed interface is that the user can input the fuzzy judgments into the system directly and easily.
However, if the user is unable to provide fuzzy comparison judgments between two elements, then he/she can click on the 'Missing Data' button and the system temporarily puts 1  for this comparison.The negative value is not a true judgment in the real world; it just indicates that those elements should    After entering the fuzzy judgments from all DMs, the user can set the DMs' importance weights into the system.According to the given data in example 1, the importance weights of the three DMs are entered, as shown in Fig. 5.
Finally, the Solver finds the optimal solution and visualises it graphically -Fig.6.

VI. CONCLUSIONS
This paper proposes a new method for solving fuzzy group prioritisation problems.The non-linear FPP is modified for group decision making by introducing DMs' importance weights.The proposed method derives crisp priorities/weights from a set of fuzzy judgements and it does not require defuzzification procedures.Moreover, the proposed method is capable of deriving crisp priorities from an incomplete set of DMs' fuzzy pairwise comparison judgments.Comparing with the Weighted FPP method, the proposed method is efficient from a computational point of view.Hence, the proposed method is a promising and attractive alternative method to existing fuzzy prioritisation methods.
Another contribution of this study is the development of a Non-Linear FPP Solver for solving group prioritisation problems, which provides a user-friendly and efficient way to obtain the group priorities.
Future work includes presenting the importance weights for the DMs as fuzzy numbers, not just as crisp numbers, in order to model the uncertain importance weights of DMs.Moreover, we would like to incorporate the proposed method into other MCDM methods such as the Fuzzy Analytic Hierarchy Process, the Fuzzy Analytic Network Process and the Evidential Reasoning approach for complex decision problem analysis.After joining the Decision Technologies Group at the Computation Department, UMIST in January 1997, Ludmil started to investigate new methods for multiple criteria decision making.He is the author of about 90 technical papers in peer-reviewed journals and international conferences, and holds two patents in the area of systems and control.His current research interests include multiple criteria decision analysis, fuzzy logic systems, decision making under uncertainty and intelligent decision support systems.Dong-Ling Xu (PhD, MBA, MEng, BEng) received her BSc degree in electrical engineering, Master degree in Business Administration (MBA), and MSc and PhD degrees in system control engineering.She is Professor of decision sciences and systems in Manchester Business School, the University of Manchester.Prior to her current appointment, she worked as lecturer and associate professor in China, principal engineer in industry and research fellow in universities in the UK.She has published over 150 papers in journals and conferences, such as European Journal of Operational Research and IEEE Transactions on Systems, Man, and Cybernetics.As a co-designer, she developed a Windows based decision support tool called IDS (Intelligent Decision Systems).The tool is now tested and used by researchers and decision analysts from over 50 countries, including organisations such as NASA, PricewaterhouseCoopers and General Motors.Her current research interests are in the areas of multiple criteria decision analysis under uncertainties, decision theory, utility theory, optimisation and their applications in performance assessment for decision making, including supplier selection, policy impact assessment, environmental impact assessment, sustainability management and consumer preference identification.

Fig. 1 .
Fig. 1.Triangular Fuzzy Number ) , , ( ~ijk ijk ijk ijk u m l a  within the scope of the initial fuzzy judgments ijk a provided by those DMs, i.e.

1
of this example, the results have been conducted by the Non-Linear FFP Solver.The solution to the non-linear problem (17) is:

Fig. 4 .
Fig. 4. The fuzzy comparison judgments window for the DM 1

Fig. 6 .
Fig. 6.The results from the Non-Linear FFP Solver does not deal with fuzzy goals; it just represents the nonlinear fuzzy constraints.Thus, by taking into account the [20]g the Non-Linear FFP Solver.It was found that the group non-linear FFP method performs significantly faster compared to the Weighted FPP[20]with different  - the third DM, who has the highest important weight, which means that the third element is about two times more important than the second element, the weights obtained by using the proposed Non-Linear FFP method are: than the first two DMs' weights, then the new fuzzy comparison judgment does not change the final ranking.Thus, we can notice the significance of introducing importance weights of the DMs to the fuzzy group prioritization problem.The computation time of the proposed method has been investigated by

TABLE III INDIVIDUAL
AND GROUP RESULTS Tarifa Almulhim received a bachelor degree in mathematics from King Faisal University, Alhassa, Saudi Arabia in 2005, and a M.Sc.degree in operational research and applied statistics from the University of Salford, Salford, UK, in 2010.She has worked as a lecturer at King Faisal University and has tutored in various subjects, including Operational Research, Applied Statistics as well as the use of SPSS software on behalf of the Statistics and Quantitative Methods Department.Currently, she is working towards her Ph.D. in business and management at the University of Manchester, UK.Her current research interests include studying the health insurance market in developing countries, multi-criteria decision methods and analytical decision processes.Ludmil Mikhailov obtained a first class BSc (1974) and MSc (1976) degree in Automatic Control from the Technical University in Sofia, Bulgaria and a PhD degree in Technical Cybernetics and Robotics from the Bulgarian Higher Certifying Commission (1981).He is a Senior Lecturer at the Manchester Business School, the University of Manchester.He worked as an Associated Professor at the Institute of Control and System Research of the Bulgarian Academy of Sciences (BAS) until 1996, where he was a head of a research group.During his work at BAS, he participated in many industrial projects on the development of various software systems for control, monitoring and technical diagnosis.