The purpose of this section is twofold. We illustrate the application of our methodology in two inspiring case-studies. We compare the results that we obtain with those of other benchmark methodologies in the same context.
7.1. An Investment Problem
In this subsection, we revisit a benchmark problem regarding venture capital investment, originally raised by Herrera and Herrera-Viedma [
54]. Wei first investigated this problem under an intuitionistic fuzzy setting in [
34]. Chen and Tu [
55] further examined this problem to demonstrate the discriminative capability of some dual bipolar measures of IFVs. Note also that Feng et al. [
51] explored the same problem to illustrate some lexicographic orders of IFVs. In what follows, this problem is to be used for illustrating Algorithm 1 in the case where the information about the attribute weights is completely unknown.
Suppose that an investment bank plans to make a venture capital investment in the most suitable company among five alternatives:
The set consisting of these companies is denoted by U, and the decision is made based on four criteria in the set , where:
represents the risk analysis;
represents the growth analysis;
represents the social-political impact analysis;
represents the environmental impact analysis.
Now, let us solve the above risk investment problem using Algorithm 1 proposed in
Section 6. The step-wise description is presented below:
Step 1. Based on the evaluation results adopted from [
34], we construct an IFSS
over
as shown in
Table 1.
Step 2. It easy to see that and since , , , and are all benefit attributes.
Step 3. By Definition 6, the generalized
-complement of
is:
and its approximate function is denoted by:
where
and
.
Step 4. In this case, the information about the attribute weights is completely unknown. Accordingly, a single-objective programming model can be established as follows:
By solving this model, we can get the following optimal solution:
After normalization, the obtained weight vector is:
Step 5. Using the weight vector
, the OIFPV
is calculated as follows:
The OIFPVs of all alternatives
(
) can be found in
Table 2.
Step 6. Assume that the decision maker specifies and the attitudinal parameter , respectively.
Step 7. Based on the selected values, we calculate the Euclidean weighted scores
of the OIFPVs according to Equation (
14):
for
. For instance, the Euclidean weighted score
can be obtained as:
The other results can be found in
Table 2.
Step 8. According to the descending order of the Euclidean weighted score
, the ranking of
(
) can be obtained as follows:
Step 9. The optimal decision is to invest in the company since it has the greatest Euclidean weighted score.
In order to verify the effectiveness of the proposed method and its consistency with existing literature, now we make a comparative check against several lexicographic orders introduced in [
51] and four other representative methods [
34,
45,
46]. We apply them to the same illustrative example that we have studied above.
According to Definition 15 and the classical scores
(
) shown in
Table 2, we can obtain the following ranking:
Table 3 summarizes all the ranking results obtained by Algorithm 1, Wei’s method [
34], Xia and Xu’s method [
45], Song et al.’s method [
46], and the lexicographic orders
,
,
, and
. By comparison, it can be seen that the ranking obtained by Algorithm 1 is only slightly different from the result of the lexicographic order
. Although the proposed method, Wei’s method [
34], Xia and Xu’s method [
45], Song et al.’s method [
46], and the lexicographic orders
,
, and
can produce the same ranking of alternatives, the rationales of these approaches is quite different.
Remark 2. Let us briefly recall some essential facts concerning the approaches that we use for contrast. Wei’s method [34] selects the weight vector w, which maximizes the total weighted deviation value among all options and with respect to all attributes. Xia and Xu’s method [45] determines the optimal weights of attributes based on entropy and cross entropy. According to Xia and Xu’s idea, an attribute with smaller entropy and larger cross entropy should be assigned a larger weight. The design of Song et al.’s similarity measure [46] for IFSs involves Jousselme’s distance measure and the cosine similarity measure between basic probability assignments (BPAs). Their similarity measure can avoid the counter-intuitive outputs by a single evaluation of the similarity measure, and it grants more rationality to the ranking results. Feng et al. [51] pointed out that lexicographic orders , , , and are not logically equivalent. Therefore, we might observe that they produce different outputs. In our case, notice that the ranking obtained under is different from the other three rankings. It is also worth noting that the rankings derived from and are identical, despite the fact that the scores are all different. This comparison analysis shows that our new approach provides an effective and consistent tool for solving multiple attribute decision-making problems under an intuitionistic fuzzy environment.
7.2. A Supplier Selection Problem
In this subsection, we illustrate Algorithm 1 in the case where the information about the attribute weights is partly known. We revisit a multi-attribute intuitionistic fuzzy group decision-making problem studied in Boran et al. [
47]. This reference analyzed a supplier selection problem by means of the TOPSIS method with intuitionistic fuzzy sets. We resort to a modified version of the case study in [
47] where the number of alternatives is artificially increased to ten, for the purpose of better understanding Algorithm 1.
Following [
47], an automotive company intends to select the most appropriate supplier for one of the key components in its manufacturing process. The selection will be made among ten alternatives in
, which are evaluated based on four criteria. The set consisting of these criteria is denoted by
, where:
stands for product quality;
stands for relationship closeness;
stands for delivery performance;
stands for product price.
Now, let us solve the above supplier selection problem using Algorithm 1. The step-wise description is given below:
Step 1. Based on the evaluation results adopted from [
47], we construct an IFSS
over
as shown in
Table 4. It should be noted that the evaluations for the alternatives
are inherited from [
47], while the evaluations of the new alternatives
are the simple intuitionistic fuzzy average (SIFA) values of the original values for
, taken by pairs.
For instance, the SIFA of the evaluations of the alternatives
and
is used as the evaluation result of the artificial alternative
. More specifically, by Equation (
5), we get:
All other evaluation results can be obtained in a similar fashion. For convenience, we simply write , , , , and .
Step 2. Divide the set C into two disjoint subsets and . It is easy to see that and , since , and are benefit attributes, while is a cost attribute.
Step 3. Calculate the generalized
-complement
of the IFSS
according to Definition 6, and the results are shown in
Table 5. For convenience, the approximate function of
is denoted by:
where
and
.
Step 4. In this case, the information about the attribute weights is partly known. Accordingly, a single-objective programming model can be established as follows:
By solving this model, we can get the following normalized optimal weight vector:
Step 5. Using the weight vector
, the OIFPVs of all alternatives
can be calculated, and the results are presented in
Table 6.
Step 6. Assume that the decision maker specifies and the attitudinal parameter , respectively. As mentioned in Remark 1, the Minkowski weighted score function is reduced to the Minkowski score function when the attitudinal parameter , representing the decision maker’s neutral attitude.
Step 7. Based on the selected values, we calculate the Minkowski weighted scores
of the OIFPVs according to Equation (
10):
for
. For instance, the Minkowski weighted score
is:
To make a thorough comparison, we also consider two other cases in which
and
, representing the decision maker’s negative and positive attitudes, respectively.
Table 6 gives all the OIFPVs and their Minkowski weighted scores with
and various choices of the attitudinal parameter
.
Step 8. The ranking of
(
) based on the descending order of the Minkowski weighted scores
can be obtained as follows:
Step 9. The optimal decision is to select the supplier since it has the greatest Minkowski weighted score.
The analysis above presumes a neutral attitude, i.e.,
. In order to visualize the influence of the attitudinal parameter on the final ranking, let us now see what conclusion we draw when we keep
and rank the alternatives with two other choices of the attitudinal parameter
.
Table 6 gives the Minkowski weighted scores
and
of the OIFPVs
. They correspond to a negative attitudinal parameter
and a positive attitude that we associate with
, respectively. It is interesting to observe that the rankings are almost identical to the ranking under a neutral attitude. This indicates that Algorithm 1 is to some extent robust with respect to the choice of the attitudinal parameter.
In addition, Boran et al. introduced the TOPSIS method based on intuitionistic fuzzy sets in [
47], which can be used for ranking IFVs as well. To further verify the rationality of the proposed methods, we proceed to compare the ranking that arises from the weight vector
obtained by Algorithm 1, the intuitionistic fuzzy TOPSIS method, and various lexicographic orders. Similarly, the ranking results of lexicographic orders can be obtained based on the classical scores
listed in
Table 6.
The attribute weight vector in [
47] is an IFV representation based on the opinions of experts, and it is different from the expression of
. Therefore, the construction of the aggregate weighted intuitionistic fuzzy decision matrix in Step 4 of the intuitionistic fuzzy TOPSIS method in [
47] must be changed. Assume that
is the fuzzy set in
C such that
(
). Then, the weighted IFSS can easily be obtained by calculating the scalar product
of the IFSS
and the fuzzy set
. For instance, by Definition 7 and Equation (
2), we have:
The weighted IFSS that arises is shown in
Table 7.
By performing the rest of the steps of the intuitionistic fuzzy TOPSIS method in [
47], we get
Table 8, which consists of the Euclidean distance
between each object and the positive ideal solution, the Euclidean distance
between each object and the negative ideal solution, and the relative closeness coefficient
. According to the intuitionistic fuzzy TOPSIS method, we can obtain the ranking results based on the descending order of the relative closeness coefficient
as shown in
Table 9.
Table 9 summarizes all the ranking results obtained by Algorithm 1 (with three different attitudinal parameters
), the intuitionistic fuzzy TOPSIS method [
47], and the lexicographic orders
,
,
, and
. By comparison, it can be seen that the ranking obtained by Algorithm 1 with the pessimistic attitude is only slightly different from those obtained with either the neutral or optimistic attitude. Algorithm 1 with either the neutral or optimistic attitude and the lexicographic orders
,
produce the same results. The ranking given by Algorithm 1 with the pessimistic attitude coincides with the results of the lexicographic orders
and the intuitionistic fuzzy TOPSIS method. Whatever the choice, the best supplier is
, and the worst supplier is
. Thus, this numerical example concerning supplier selection further illustrates that the method proposed in this study is feasible for solving multiple attribute decision-making problems in real-life environments.
Remark 3. Let us briefly highlight several key points with regard to the above discussion. Note first that the set of alternatives is expanded from the original set [47] consisting of five alternatives to a new set containing ten alternatives. All the evaluation results of these alternatives are shown in Table 4. It should be noted that the evaluation results of the alternatives () are inherited directly from [47]. In contrast, the alternatives () in Table 4 are artificial ones, and the evaluation results of them are obtained by computing the SIFA values of the IFVs associated with in a pairwise manner. Note also that the obtained artificial alternatives provide an additional way to verify the rationality of Algorithm 1. In fact, we can see from Table 9 that each artificial alternative lies between the two original alternatives that are used to produce it. For instance, lies between and in all the ranking results. This is due to the fact that we obtain the evaluation results of by simply averaging the corresponding results of and . Finally, the ranking results in Table 9 with , , and show that the selection of different attitudinal parameters may affect the final ranking results. However, to a certain extent, Algorithm 1 is robust with respect to the choice of the attitudinal parameter as well.