Delphi Method for Estimating Membership Function of Uncertain Set
 Haiying Guo^{1},
 Xiaosheng Wang^{1}Email author,
 Lili Wang^{1} and
 Dan Chen^{1}
DOI: 10.1186/s4046701600441
© Guo et al. 2016
Received: 9 November 2015
Accepted: 23 January 2016
Published: 9 February 2016
Abstract
Uncertain set is a setvalued function on an uncertainty space, and attempts to model unsharp concepts. Membership function is used to describe the membership degree that a value belongs to the uncertain set. In order to estimate the membership function for an uncertain set via multiple experts’ experimental data, this paper presents a new method by the combination of the method of uncertain statistics and Delphi method, and gives an example to verify the method.
Keywords
Uncertain set Membership function Uncertain statistics Delphi methodIntroduction
In the dance competition, experts give their own judgments such as superior, middle, or inferior according to the dancers’ performance. For the examination paper, teachers tend to judge the degrees of papers’ difficulty in advance. In fact, when forecasting and decisionmaking in practice, experts will judge according to their own experience and knowledge. And, experts’ advice is often crucial. How to distinguish the experience and knowledge given by experts? How to understand experts’ experience and knowledge? How to express in mathematical language? Different scholars have proposed the different approaches. A traditional method is to use probability theory. The experts’ advice is characterized by subjective probability. Another method is to use the fuzzy set proposed by Zadeh [1]. Experts’ experience and knowledge are interpreted as experts’ belief degree by the two methods in common. Each method is constantly challenged since being proposed. It is pointed that human beings usually overweight are unlikely events by Kahneman and Tversky [2], so the belief degree may have much larger variance than the real frequency. Now, if we deal with the belief degree by probability theory, it will lead to making wrong decisions. A specific counterexample supporting this view was presented by Liu [3]. Another example was also presented by Liu [3] saying that the fuzzy set is not suitable for unsharp concepts such as “young”.
In order to model human uncertainty well with the axiomatic mathematical model, uncertainty theory was found by Liu [4] in 2007 and refined by Liu [5] in 2010. The basic idea of the uncertainty theory comes from the measure theory in classical mathematics and probability theory. The first fundamental concept in the uncertainty theory is an uncertain measure. It is used to indicate the belief degree that an uncertain event may occur. The second one is uncertain variable that represents quantities in uncertainty. The third one is uncertainty distribution to describe an uncertain variable. In order to deal with expert’s experimental data by uncertainty theory, Liu [6] first built the uncertain statistical method in 2010. Uncertain statistics is based on experts’ experimental data collected by designing a questionnaire survey. In uncertain statistics, Liu [6] firstly suggested an empirical uncertainty distribution and proposed a principle of least squares as the method for estimating the unknown parameters. In addition, Wang and Peng [7] proposed a moment method for estimating the unknown parameters in uncertain statistics in 2010. Also, Wang et al. [8] presented uncertain hypothesis testing to detect whether two uncertainty distributions are equal. In order to estimate the uncertainty distribution for an uncertain variable via multiple experts’ experimental data, Wang et al. [9] proposed a new method based on the uncertain statistic and Delphi method.
As one of the main contents of uncertainty theory, the concept of uncertain set was first proposed by Liu [10] in 2010 and recast by Liu [11] in 2012. The uncertain set is used to model “unsharp concepts” like “young”, “tall”, “most”, and “about 100 km”. The expected value of the uncertain set was defined by Liu [10]. Liu [12] introduced the concepts of variance, entropy, and distance of uncertain sets. More importantly, Liu [11] presented the concept of membership function to describe the uncertain set and provided the operational law of uncertain sets via membership function or inverse membership function. Based on the uncertain set theory, uncertain logic was designed by Liu [12] for dealing with human language by using the truth value formula for uncertain propositions. In addition, a basic uncertain inference rule was proposed by Liu [10]. After that, Gao et al. [13] extended the inference rule to the case with multiple antecedents and multiple ifthen rules. Based on the uncertain inference rule, Liu [10] suggested a concept of the uncertain system and then presented an uncertain inference controller as a tool. As a contribution, Peng and Chen [14] proved that an uncertain system is a universal approximator. As a successful application, Gao [15] balanced an inverted pendulum by using the uncertain inference controller.
The uncertain set is mainly characterized by the membership function and operation law. Now, the problem we face is how to get the membership function and how to gain experts’ experimental data. In order to determine the membership function of an uncertain set, Liu [12] designed a questionnaire survey to collect experts’ experimental data and suggested an empirical membership function. Liu [12] also proposed a principle of least squares as a method for estimating the unknown parameters based on the expert’s experimental data. In this paper, we propose a new method aiming at determining the membership function by the combination of the method of uncertain statistics and Delphi method.
The rest of this paper is organized as follows. The next section is intended to introduce some concepts in uncertainty theory as they are needed. Some basic concepts of uncertain statistics are introduced in Section “Experts’ Experimental Data”. A new method is presented to determine membership function in Section “A New Method”. An example is proposed in Section “Estimating the Age Range of the Uncertain Set “young people”” to test the method. Finally, a conclusion is drawn in Section “Conclusions”.
Preliminaries
In this section, we will introduce some useful definitions about the uncertain measure, uncertain set, membership function and so on.
Let Γ be a nonempty set, and be a σalgebra over Γ. Each element is called an event. A number indicates the level that Λ will occur. Uncertain measure was introduced as a set function satisfying the following axioms (Liu [4]):
Axiom 1.
(Normality Axiom) for the universal set Γ.
Axiom 2.
(Duality Axiom) for any event Λ.
Axiom 3.
(Subadditivity Axiom) For every countable sequence of events {Λ _{ i }}, we have
The triplet is called an uncertainty space. In addition, Liu [16] defined the product uncertain measure as follows.
Axiom 4.
(Product Axiom) Let be uncertainty spaces for k=1,2,⋯ The product uncertain measure is an uncertain measure satisfying
where Λ _{ k } are arbitrarily chosen events from for k=1,2,⋯, respectively.
The concept of uncertain variable ξ was introduced by Liu [4] as a measurable function from an uncertainty space to the set of real numbers. In order to describe an uncertain variable, uncertainty distribution is defined by Liu [4] as
Furthermore, the inverse uncertain distribution Φ ^{−1}(α) of ξ was defined by Liu [3]. It plays a crucial role in operations of an uncertain variable.
The expected value is the average value of uncertain variable in the sense of the uncertain measure, and represents the size of the uncertain variable. It was defined by Liu [4] as follows,
Different from the uncertain variable that is a realvalued function, the uncertain set is a setvalued function on an uncertainty space. A formal definition was given by Liu [10] as follows.
Definition 1.
In order to facilitate the operations, Liu defined the independence of the uncertain set as follows.
Definition 2.
(Liu [17]) The uncertain sets ξ _{1},ξ _{2},⋯,ξ _{ n } are said to be independent if for any Borel set B _{1},B _{2},⋯,B _{ n }, we have
and
where \(\xi ^{*}_{i}\) are arbitrarily chosen from \(\left \{\xi _{i},{\xi ^{c}_{i}}\right \}, i=1,2,\cdots,n,\) respectively.
Membership function was proposed by Liu to describe the uncertain set.
Definition 3.
(Liu [11]) An uncertain set ξ is said to have a membership function μ if for any Borel set B of real numbers, we have
Theorem 1.
To make the operation easy, Liu proposed the definition of the inverse membership function of the uncertain set and gave some operation laws of independent uncertain sets.
Definition 4.
Sometimes, the set μ ^{−1}(α) is also called the αcut of μ.
Theorem 2.
Theorem 3.
Theorem 4.
Theorem 5.
Remark 1.
The above arithmetic operational law is not equivalent to the extension principle of Zadeh although they coincide with each other in many cases.
Experts’ Experimental Data
Uncertain statistics is based on experts’ experimental data rather than historical data. In order to determine the membership function of the uncertain set, Liu [12] designed a questionnaire survey for collecting experts’ experimental data. The starting point is to invite one or more domain experts who are asked to complete a questionnaire about the concept of an uncertain set ξ like “about young” individually.
A New Method
Assume that there are m domain experts and each produces an empirical membership function. Then, we may get m empirical membership functions μ _{1}(x),μ _{2}(x),⋯,μ _{ m }(x).
Theorem 6.
is a membership function, where ω _{ i }≥0,i=1,2,⋯,m and \(\sum \limits _{i=1}^{m}\omega _{i}=1\).
Proof.
Since μ _{1}(x),μ _{2}(x),⋯,μ _{ m }(x) are membership functions, by Theorem 1, we have 0≤μ _{ i }(x)≤1,i=1,2,⋯,m. Since ω _{ i }≥0,i=1,2,⋯,m and \(\sum \limits _{i=1}^{m}\omega _{i}=1\), then we have \(0\leq \mu (x)=\sum \limits _{i=1}^{m}\omega _{i}\mu _{i}(x)\leq 1\). Thus μ(x) is also a membership function by Theorem 1. □
Remark 2.
ω _{ i },i=1,2,⋯,m are convex combination coefficients representing weighs.
Since its inceptive development by Dalkey and Helmer [18, 19], the Delphi method has been widely accepted as an effective forecasting tool and used in a wide range of applications. It is used to structure a group communication process to facilitate group problem solving and to structure models [20]. The method can also be used as a judgement, decisionaiding, or forecasting tool [21], and can be applied to program planning and administration [22]. The Delphi method can be used when there is incomplete knowledge about a problem or phenomena [22, 23].
The steps of the Delphi method are as follows. Firstly, the participants assess the matters independently and make individual judgement according to individual knowledge and experience. Then, from the second round, the participants are provided with feedback on the previous round so that they can assess the same matters again and make a new judgement about altering their opinions, repeating the above process. Finally, the experts’ opinions will tend to be consistent, and a composite survey result is presented.

Step 1. Respectively, the m domain experts provide their experimental data (x _{ ij },α _{ ij }), where x _{ ij } denotes the jth possible value provided by the ith expert and α _{ ij } denotes the ith expert’s belief degree that x _{ ij } belongs to ξ, i=1,2,⋯,m, j=1,2,⋯,n _{ i }, respectively.

Step 2. Use the ith expert’s experimental data \(\phantom {\dot {i}\!}(x_{i1},\alpha _{i1}), (x_{i2},\alpha _{i2}), \cdots, (x_{in_{i}},\alpha _{in_{i}})\) to generate the empirical membership function μ _{ i } of the ith domain experts, i=1,2,⋯,m, respectively.

Step 3. We calculate the number of the possible values of ξ presented by all experts and denote it by n, where the same values from different experts are treated as one. Then, the possible values of ξ are x _{1}≤x _{2}≤⋯≤x _{ n }, and we continue to computeand$$\bar{\alpha}_{j}=\frac{1}{m}\sum_{i=1}^{m}\mu_{i}(x_{j}),\ j=1,2,\cdots,n$$$$d_{j}=\frac{1}{m}\sum_{i=1}^{m}(\mu_{i}(x_{j})\bar{\alpha}_{j})^{2},\ j=1,2,\cdots,n. $$

Step 4. If d _{ j } are less than a given level ε>0 for all j, then go to Step 5. Otherwise, the ith domain expert receives the summary (for example, the \(\bar {\alpha }_{j}\) obtained in the previous round and the reasons of other experts), and then provide a set of revised experts’ experimental data \(\phantom {\dot {i}\!}(x_{i1},\alpha _{i1}), (x_{i2},\alpha _{i2}), \cdots, (x_{in_{i}},\alpha _{in_{i}})\) for i=1,2,⋯,m, go to Step 2.

Step 5. Use the integrated data \((x_{1}, \bar {\alpha }_{1}), (x_{2}, \bar {\alpha }_{2}), \cdots, (x_{n}, \bar {\alpha }_{n})\) to generate the membership function μ(x) of the uncertain set ξ.
Estimating the Age Range of the Uncertain Set “young people”
In this section, we will give an example to verify our method that combines uncertain statistics and the Delphi method for estimating the membership function.
The uncertain set is a setvalued function on an uncertainty space and attempts to model “unsharp concepts” that are essentially sets but their boundaries are not sharply described (because of the ambiguity of human language). The concepts that being an uncertain set do not have the property of exclusivity.
where E _{ i } represents the ith expert, i=1,2,⋯,6, respectively.
The first round belief degree
x  15  16  17  18  19  20  21  22  24  25  26  27  28  29  30  31  32  34  35 

\(\mu _{1}^{(1)}(x)\)  0.0000  0.0000  0.0000  0.0000  0.3500  0.7000  0.9000  1.0000  1.0000  1.0000  1.0000  0.9000  0.6000  0.3000  0.0000  0.0000  0.0000  0.0000  0.0000 
\(\mu _{2}^{(1)}(x)\)  0.0000  0.0000  0.0000  0.7500  0.8250  0.9000  1.0000  1.0000  1.0000  0.8500 0.7000  0.6000  0.5000  0.2500  0.0000  0.0000  0.0000  0.0000  0.0000  
\(\mu _{3}^{(1)}(x)\)  0.0000  0.0000  0.0000  0.0000  0.7000  0.8000  0.9000  1.0000  1.0000  0.5000  0.3500  0.2000  0.0000  0.0000  0.0000  0.0000  0.0000  0.0000  0.0000 
\(\mu _{4}^{(1)}(x)\)  0.0000  0.0000  0.0000  0.2500  0.5000  0.9000  1.0000  1.0000  1.0000  1.0000  0.8000  0.6000  0.4000  0.2000  0.0000  0.0000  0.0000  0.0000  0.0000 
\(\mu _{5}^{(1)}(x)\)  0.0000  0.3000  0.6000  0.9000  0.9500  1.0000  1.0000  1.0000  1.0000  1.0000  0.9500  0.9000  0.8500  0.7750  0.7000  0.5000  0.3000  0.1000  0.0000 
\(\mu _{6}^{(1)}(x)\)  0.0000  0.0000  0.4000  0.8000  0.9000  1.0000  1.0000  1.0000  1.0000  1.0000  1.0000  1.0000  1.0000  1.0000  1.0000  0.7000  0.6333  0.5000  0.0000 
The first round mean values and deviations
x  15  16  17  18  19  20  21  22  24  25  26  27  28  29  30  31  32  34  35 

\(\bar {\alpha }_{j}^{(1)}\)  0.0000  0.0500  0.1667  0.4500  0.7042  0.8833  0.9667  1.0000  1.0000  0.8917  0.8000  0.7000  0.5583  0.4208  0.2833  0.2000  0.1556  0.1000  0.0000 
\(d_{j}^{(1)}\)  0.0000  0.0125  0.0589  0.1433  0.0468  0.0114  0.0022  0.0000  0.0000  0.0337  0.0525  0.0733  0.1037  0.1218  0.1681  0.0833  0.0577  0.0333  0.0000 
The second round belief degree
x  15  16  17  18  19  20  21  22  24  25  26  27  28  29  30  31  32  34  35 

\(\mu _{1}^{(2)}(x)\)  0.0000  0.0000  0.0000  0.2000  0.5500  0.9000  1.0000  1.0000  0.8500  0.7750  0.7000  0.6000  0.5000  0.3000  0.0000  0.0000  0.0000  0.0000  0.0000 
\(\mu _{2}^{(2)}(x)\)  0.0000  0.0000  0.1000  0.5000  0.7000  0.9000  1.0000  1.0000  1.0000  0.8500  0.7000  0.6000  0.5000  0.3000  0.0000  0.0000  0.0000  0.0000  0.0000 
\(\mu _{3}^{(2)}(x)\)  0.0000  0.0000  0.0000  0.3000  0.8000  0.9000  0.9500  1.0000  1.0000  0.9000  0.8500  0.8000  0.0000  0.0000  0.0000  0.0000  0.0000  0.0000  0.0000 
\(\mu _{4}^{(2)}(x)\)  0.0000  0.0000  0.2000  0.5000  0.8000  0.9000  1.0000  1.0000  1.0000  1.0000  0.9000  0.8000  0.6500  0.5000  0.0000  0.0000  0.0000  0.0000  0.0000 
\(\mu _{5}^{(2)}(x)\)  0.0000  0.2000  0.4000  0.6000  0.8000  1.0000  1.0000  1.0000  1.0000  1.0000  0.9000  0.8000  0.7000  0.5500  0.4000  0.3000  0.2000  0.0667  0.0000 
\(\mu _{6}^{(2)}(x)\)  0.0000  0.1000  0.3000  0.6000  0.8000  1.0000  0.9400  0.8800  0.7600  0.7000  0.6400  0.5800  0.5200  0.4600  0.4000  0.3000  0.2667  0.2000  0.0000 
The second round mean values and deviations
x  15  16  17  18  19  20  21  22  24  25  26  27  28  29  30  31  32  34  35 

\(\bar {\alpha }_{j}^{(2)}\)  0.0000  0.0500  0.1667  0.4500  0.7417  0.9333  0.9817  0.9800  0.9350  0.8708  0.7817  0.6967  0.4783  0.3517  0.1333  0.1000  0.0778  0.0444  0.0000 
\(d_{j}^{(2)}\)  0.0000  0.0058  0.0222  0.0225  0.0087  0.0022  0.0007  0.0020  0.0091  0.0122  0.0110  0.0107  0.0517  0.0337  0.0356  0.0200  0.0125  0.0054  0.0000 
The third round belief degree
x  15  16  17  18  19  20  21  22  24  25  26  27  28  29  30  31  32  34  35 

\(\mu _{1}^{(3)}(x)\)  0.0000  0.0000  0.0000  0.5000  0.7000  0.9000  1.0000  1.0000  0.9000  0.8500  0.8000  0.7000  0.5000  0.3500  0.0000  0.0000  0.0000  0.0000  0.0000 
\(\mu _{2}^{(3)}(x)\)  0.0000  0.0000  0.2000  0.5000  0.7000  0.9000  1.0000  0.9667  0.9000  0.8500  0.8000  0.6500  0.5000  0.3500  0.0000  0.0000  0.0000  0.0000  0.0000 
\(\mu _{3}^{(3)}(x)\)  0.0000  0.0000  0.0000  0.5000  0.8000  0.9000  0.9500  1.0000  1.0000  0.9000  0.8000  0.7000  0.0000  0.0000  0.0000  0.0000  0.0000  0.0000  0.0000 
\(\mu _{4}^{(3)}(x)\)  0.0000  0.0000  0.2000  0.5000  0.8000  0.9000  1.0000  0.9750  0.9250  0.9000  0.8000  0.7000  0.5500  0.4000  0.0000  0.0000  0.0000  0.0000  0.0000 
\(\mu _{5}^{(3)}(x)\)  0.0000  0.2000  0.4000  0.6000  0.8000  1.0000  0.9800  0.9600  0.9200  0.9000  0.7667  0.6333  0.5000  0.3500  0.2000  0.1500  0.1000  0.0333  0.0000 
\(\mu _{6}^{(3)}(x)\)  0.0000  0.1000  0.2000  0.6000  0.8000  1.0000  0.9200  0.8400  0.6800  0.6000  0.5200  0.4400  0.3600  0.2800  0.2000  0.1000  0.0667  0.0000  0.0000 
The third round mean values and deviations
x  15  16  17  18  19  20  21  22  24  25  26  27  28  29  30  31  32  34  35 

\(\bar {\alpha }_{j}^{(3)}\)  0.0000  0.0500  0.1667  0.5333  0.7667  0.9333  0.9750  0.9569  0.8875  0.8333  0.7478  0.6372  0.4017  0.2883  0.0667  0.0417  0.0278  0.0056  0.0000 
\(d_{j}^{(3)}\)  0.0000  0.0058  0.0189  0.0022  0.0022  0.0022  0.0009  0.0030  0.0097  0.0114  0.0105  0.0085  0.0357  0.0178  0.0089  0.0037  0.0016  0.0002  0.0000 
(15, 0), (16, 0.0500), (17, 0.1667), (18, 0.5333), (19, 0.7667), (20, 0.9333), (21, 0.9750), (22, 0.9569), (24, 0.8875), (25, 0.8333), (26, 0.7478), (27, 0.6372), (28, 0.4017), (29, 0.2883), (30, 0.0667), (31, 0.0417), (32, 0.0278), (34, 0.0056), and (35, 0).
We consider people whose ages’ membership α>0.5 are “young people”. According to Fig. 1, we know the range of young people’s ages is 18–27.
Conclusions
Uncertain statistics is a methodology for collecting and interpreting experts’ experimental data by uncertainty theory. The method in this paper is a new way to get the empirical membership function when dealing with multiple experts’ experimental data. This method depends on each expert’s membership function and the Delphi method, and the example shows that this method is effective.
Declarations
Acknowledgements
This work was supported by Hebei Natural Science Foundations No.G2013402063 and Hebei Social Science Foundation (Grant No. HB14JY068).
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.
Authors’ Affiliations
References
 Zadeh, L: Fuzzy sets. Inform. Control. 8, 338–353 (1965).View ArticleMathSciNetMATHGoogle Scholar
 Kahneman, D, Tversky, A: Prospect theory: an analysis of decision under risk. Econometrica. 47(2), 263–292 (1979).View ArticleMATHGoogle Scholar
 Liu, B: Why is there a need for uncertainty theory?J. Uncertain Syst. 6(1), 3–10 (2012).Google Scholar
 Liu, B: Uncertainty Theory. 2nd ed. SpringerVerlag, Berlin (2007).View ArticleMATHGoogle Scholar
 Liu, B: Uncertainty theory: a branch of mathematics for modeling human uncertainty. SpringerVerlag, Berlin (2010).View ArticleGoogle Scholar
 Liu, B: Uncertainty theory. 5th ed (2015). http://orsc.edu.cn/liu/ut.pdf.
 Wang, X, Peng, Z: Method of moments for estimating uncertainty distributions. J. Uncertainty Anal. Appl. 2(5), 1–11 (2014).View ArticleGoogle Scholar
 Wang, X, Gao, Z, Guo, H: Uncertain hypothesis testing for two experts’ empirical data. Math. Comput. Modell. 55, 1478–1482 (2012).View ArticleMathSciNetMATHGoogle Scholar
 Wang, X, Gao, Z, Guo, H: Delphi method for estimating uncertainty distributions. Inform. Int. Interdiscip. J. 15(2), 449–460 (2012).MathSciNetGoogle Scholar
 Liu, B: Uncertain set theory and uncertain inference rule with application to uncertain control. J. Uncertain Syst. 4(2), 83–98 (2010).Google Scholar
 Liu, B: Membership functions and operational law of uncertain sets. Fuzzy Optimization Decis. Making. 11(4), 387–410 (2012).View ArticleMATHGoogle Scholar
 Liu, B: Uncertain logic for modeling human language. J. Uncertain Syst. 5(1), 3–20 (2011).Google Scholar
 Gao, X, Gao, Y, Ralescu, DA: On Liu’s inference rule for uncertain systems. Int. J. Uncertainty, Fuzziness KnowledgeBased Syst. 18(1), 1–11 (2010).View ArticleMathSciNetMATHGoogle Scholar
 Peng, Z, Chen, X: Uncertain systems are universal approximators. J. Uncertainty Anal. Appl. 2(13), 1–10 (2014).MathSciNetGoogle Scholar
 Gao, Y: Uncertain inference control for balancing an inverted pendulum. Fuzzy Optimization Decis. Making. 11(4), 481–492 (2012).View ArticleMATHGoogle Scholar
 Liu, B: Some research problems in uncertainty theory. J. Uncertain Syst. 3(1), 3–10 (2009).Google Scholar
 Liu, B: A new definition of independence of Uncertain sets. Fuzzy Optimization Decis. Making. 12(4), 451–461 (2013).View ArticleGoogle Scholar
 Norman, D, Olaf, H: An experimental application of the Delphi method to the use of experts. Manag. Sci. 9(458) (1963).
 Olaf, H: The Delphi method for systematizing judgments about the future. Institute of Government and Public Affairs, University of California (1966).
 Linstone, H, Turoff, M: The Delphi Method. AddisonWesley Publishing Company, London (1975).MATHGoogle Scholar
 Rowe, G, Wright, G: The Delphi technique as a forecasting tool: issues and analysis. Int. J. Forecasting. 15(4), 353–375 (1999).View ArticleGoogle Scholar
 Delbeq, A, Van de Ven, A, Gustafson, DH: Group techniques for program planning: a guide to nominal group and Delphi processes (1975).
 Adler, M, Ziglio, E: Gazing into the oracle: The Delphi Method and its application to social policy and public health (1996). Publishers.