Sei sulla pagina 1di 13

Information Sciences 255 (2014) 45–57

Contents lists available at ScienceDirect

Information Sciences
journal homepage: www.elsevier.com/locate/ins

A biparametric similarity measure on intuitionistic fuzzy sets


with applications to pattern recognition
Fatih Emre Boran, Diyar Akay ⇑
Department of Industrial Engineering, Faculty of Engineering, Gazi University, 06570 Ankara, Turkey

a r t i c l e i n f o a b s t r a c t

Article history: Unlike an ordinary fuzzy set, the concept of intuitionistic fuzzy set (IFS), characterized both
Received 25 October 2011 by a membership degree and by a non-membership degree, is a more flexible way to cap-
Received in revised form 6 August 2013 ture the uncertainty. One of the important topics in IFS is the measure of the similarity
Accepted 7 August 2013
between IFSs for which several studies have been proposed in the literature. Some of those,
Available online 23 August 2013
however, cannot satisfy the axioms of similarity, and provide counter-intuitive cases. In
this paper, a new general type of similarity measure for IFS with two parameters is pro-
Keywords:
posed along with its proofs. A comparison between the existing similarity measures and
Intuitionistic fuzzy set
Distance measure
the proposed similarity measure is also performed in terms of counter-intuitive cases.
Similarity measure The findings indicate that the proposed similarity measure does not provide any coun-
Pattern recognition ter-intuitive cases.
Ó 2013 Elsevier Inc. All rights reserved.

1. Introduction

The theory of fuzzy set, proposed by Zadeh [1], has received a great deal of attention due to its capability of handling
uncertainty. Therefore, over the last decades, several higher order fuzzy sets have been introduced in the literature. Intui-
tionistic fuzzy set (IFS), proposed by Atanassov [2], is one of the higher order fuzzy sets which is capable of dealing with
vagueness. An IFS is characterized by three parameters, namely a membership degree, a non-membership degree, and a hes-
itation margin, while a fuzzy set is characterized by only a membership degree. IFS is therefore a more effective way to deal
with vagueness than fuzzy set. Although Gau and Buehrer [3] later presented vague set, Bustince and Burillo [4] pointed out
that the notion of vague sets was the same as that of IFS.
The degree of similarity measure has received a great deal of attention in the last decades since it is an important tool for
decision making, pattern recognition, medical diagnosis, and the applications of data mining [5]. For that reason, some stud-
ies on the measure of similarity between IFSs have been reported in the literature. A few of them is the extension of the well-
known distance measures. The first study was carried out by Szmidt and Kacprzyk [6] extending the well-known distances
measures, such as the Hamming Distance, the Euclidian Distance, to IFS environment and comparing them with the ap-
proaches used for ordinary fuzzy sets. However, Wang and Xin [7] implied that the distance measure of Szmidt and Kacprzyk
[6] were not effective in some cases. Therefore, several new distance measures were proposed and applied to pattern recog-
nition. Grzegorzewski [8] also extended the Hamming distance, the Euclidean distance, and their normalized counterparts to
IFS environment. Later, Chen [9] pointed out that some errors existed in Grzegorzewski [8] by showing some counter exam-
ples. Hung and Yang [10] extended the Hausdorff distance to IFSs and proposed three similarity measures. On the other hand,
instead of extending the well-known measures, some studies defined new similarity measures for IFSs. Li and Cheng [11]
suggested a new similarity measure for IFSs based on the membership degree and the non-membership degree. Mitchell

⇑ Corresponding author. Tel.: +90 312 5823844; fax: +90 312 2308434.
E-mail address: diyar@gazi.edu.tr (D. Akay).

0020-0255/$ - see front matter Ó 2013 Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.ins.2013.08.013
46 F.E. Boran, D. Akay / Information Sciences 255 (2014) 45–57

[12] showed that the similarity measure of Li and Cheng [11] had some counter-intuitive cases and modified that similarity
measure based on statistical point of view. Moreover, Liang and Shi [13] presented some examples to show that the simi-
larity measure of Li and Cheng [11] was not reasonable for some conditions, and therefore proposed several new similarity
measures for IFSs. Li et al. [14] analyzed, compared and summarized the existing similarity measures between IFSs/vague
sets by their counter-intuitive examples in pattern recognition. Ye [15] conducted a similar comparative study of the existing
similarity measures between IFSs and proposed a cosine similarity measure and a weighted cosine similarity measure.
Hwang et al. [16] proposed a similarity measure for IFSs in which Sugeno integral was used for aggregation. The proposed
similarity measure was applied to clustering problem. Xu [17] introduced a series of similarity measures for IFSs and applied
them to multiple attribute decision making problem based on intuitionistic fuzzy information. Xu and Chen [18] introduced
a series of distance and similarity measures, which are various combinations and generalizations of the weighted Hamming
distance, the weighted Euclidean distance and the weighted Hausdorff distance. Xu and Yager [19] developed a similarity
measure between IFSs and applied the developed similarity measure for consensus analysis in group decision making based
on intuitionistic fuzzy preference relations. Xia and Xu [5] proposed a series of distance measures based on the intuitionistic
fuzzy point operators. In addition to these studies, some works have been interested in relationships between distance mea-
sure, similarity measure and entropy of IFSs. Zeng and Guo [20] investigated the relationship among the normalized dis-
tance, the similarity measure, the inclusion measure, and the entropy of interval-valued fuzzy sets. It was also showed
that the similarity measure, the inclusion measure, and the entropy of interval-valued fuzzy sets could be induced by the
normalized distance of interval-valued fuzzy sets based on their axiomatic definitions. Wei et al. [21] introduced a entropy
measure generalizing the existing entropy measures for IFS and IFSs. It was also introduced an approach to construct sim-
ilarity measures using entropy measures for IFS and IFSs.
In this paper, we introduce a new distance measure between IFSs and give its relation with the similarity measure for IFSs.
The proposed generalized distance measure on intuitionistic fuzzy sets be presented in Eq. (8) depends on two parameters
where p is the Lp norm and t identifies the level of uncertainty. We compare the existing similarity measures with the pro-
posed similarity measure for IFSs. In order to do so, the rest of this paper is organized as follows. Section 2 presents the def-
initions related to the IFSs, and lists the properties that a distance measure for IFSs and a similarity measure for IFSs should
possess. The new distance measure and corresponding novel type of similarity measure are expressed in Section 3. The inter-
pretation of new distance measure and the explanation of its parameter are briefly introduced in Section 4. A comparative
analysis between the proposed similarity measure and the existing similarity measures is presented in Section 5. The appli-
cations of the proposed similarity measure to pattern recognition are presented in Section 6. The conclusion of the paper is
given in Section 7.

2. Preliminaries

In this section, we briefly introduce the basic concepts related to IFS, and then list the properties that a distance measure
for IFSs and a similarity measure for IFSs should possess.

Definition 1 [1]. A fuzzy set A in the universe of discourse X = {x1,x2, . . . , xn} is defined as follows:
A ¼ fhx; lA ðxÞijx 2 Xg ð1Þ
where lA(x): X ? [0, 1] is the membership degree.

Definition 2 [2]. An IFS A in a finite set X can be written as:


A ¼ fhx; lA ðxÞ; v A ðxÞijx 2 Xg ð2Þ
where lA(x): X ? [0, 1] and vA(x): X ? [0, 1] are membership degree and non-membership degree, respectively, such that:
0 6 lA ðxÞ þ v A ðxÞ 6 1 ð3Þ
The third parameter of the IFS is:
pA ðxÞ ¼ 1  lA ðxÞ  v A ðxÞ ð4Þ
which is known as the intuitionistic fuzzy index or the hesitation degree of whether x belongs to A or not. It is obviously seen
that for every:
0 6 pA ðxÞ 6 1 ð5Þ
If pA(x) is small, then knowledge about x is more certain; if pA(x) is great, then knowledge about x is more uncertain. Obvi-
ously, when vA(x) = 1  lA(x) for all elements of the universe, the ordinary fuzzy set is recovered [22].

~ ¼ ðla ; v a Þ be an intuitionistic fuzzy number (IFN), then the score function of a


Definition 3. Let a ~ where sða
~Þ 2 ½1; 1 is
defined as follows [23]:
F.E. Boran, D. Akay / Information Sciences 255 (2014) 45–57 47

~Þ ¼ ðla  v a Þ
sða ð6Þ
~ where hða
and the accuracy function of a ~Þ 2 ½0; 1 is defined as follows [24]:
~Þ ¼ ðla þ v a Þ
hða ð7Þ
~1 and a
Let a ~2 be two IFNs, then Xu and Yager [25] proposed the following rules for ranking of IFNs:

1. sða~1 Þ < sða~2 Þ then a ~1 is smaller than a~2 , denoted by a ~1 < a


~2 .
2. sða~1 Þ > sða~2 Þ then a ~2 is smaller than a~1 , denoted by a ~1 > a
~2 .
3. sða~1 Þ ¼ sða~2 Þ then
(i) hða ~1 Þ ¼ hða ~2 Þ then a~1 is equal to a
~2 , denoted by a ~1 ¼ a
~2 .
(ii) hða ~1 Þ < hða ~2 Þ then a~1 is smaller than a ~2 , denoted by a
~1 < a~2 .
(iii) hða ~1 Þ > hða ~2 Þ then a~2 is smaller than a ~1 , denoted by a
~1 > a~2 .

Definition 4. A mapping D: IFS  IFS ? [0, 1], D(A, B) is said to be a distance between A 2 IFS(X) and B 2 IFS(X) if D(A, B) sat-
isfies the following properties:

(A1) 0 6 D(A, B) 6 1
(A2) D(A, B) = 0 if and only if A = B
(A3) D(A, B) = D(B, A)
(A4) If A # B # C then D(A, C) P D(A, B)

Definition 5. A mapping S: IFS  IFS ? [0, 1], S(A, B) is said to be a degree of similarity between A 2 IFS(X) and B 2 IFS(X) if
S(A, B) satisfies the following properties:

(A5) 0 6 S(A, B) 6 1
(A6) S(A, B) = 1 if and only if A = B
(A7) S(A, B) = S(B, A)
(A8) If A # B # C then S(A, C) 6 S(A, B)

3. A new type of distance measure between intuitionistic fuzzy sets

In this section, we propose a new distance measure. Let A and B be two IFSs in X where X = {x1, x2, . . . , xn}.
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 Xn
fjtðlA ðxi Þ  lB ðxi ÞÞ  ðv A ðxi Þ  v B ðxi ÞÞjp þ jtðv A ðxi Þ  v B ðxi ÞÞ  ðlA ðxi Þ  lB ðxi ÞÞjp g
p
DðA; BÞ ¼ p
2nðt þ 1Þ i¼1 ð8Þ
where t ¼ 2; 3; 4; . . . and p ¼ 1; 2; 3; . . .
Here, two parameters, p is the Lp norm and t identifies the level of uncertainty, are expressed in detail in Section 4.

Theorem 1. D(A, B) is the distance between two IFSs A and B in X.

Proof 1.

A(1) Let A and B be two IFSs.

We can write the following equations:

jtðlA ðxi Þ  lB ðxi ÞÞ  ðv A ðxi Þ  v B ðxi ÞÞj ¼ jðt lA ðxi Þ  v A ðxi ÞÞ  ðtlB ðxi Þ  v B ðxi ÞÞj ð9Þ
jtðv A ðxi Þ  v B ðxi ÞÞ  ðlA ðxi Þ  lB ðxi ÞÞj ¼ jðt v A ðxi Þ  lA ðxi ÞÞ  ðtv B ðxi Þ  lB ðxi ÞÞj ð10Þ
We know that 0 6 lA(xi) 6 1, 0 6 lB(xi) 6 1, 0 6 vA(xi) 6 1 and 0 6 vB(xi) 6 1, and therefore we have the following
inequalities:

 1 6 ðt lA ðxi Þ  v A ðxi ÞÞ 6 t ð11Þ


 t 6 ðtlB ðxi Þ  v B ðxi ÞÞ 6 1 ð12Þ
48 F.E. Boran, D. Akay / Information Sciences 255 (2014) 45–57

then we have
ðt þ 1Þ 6 ðt lA ðxi Þ  v A ðxi ÞÞ  ðtlB ðxi Þ  v B ðxi ÞÞ 6 ðt þ 1Þ ð13Þ
It means that
0 6 jtðlA ðxi Þ  lB ðxi ÞÞ  ðv A ðxi Þ  v B ðxi ÞÞjp 6 ðt þ 1Þp ð14Þ
Similarly, we have the following inequalities:
 1 6 ðt v A ðxi Þ  lA ðxi ÞÞ 6 t ð15Þ
 t 6 ðt v B ðxi Þ  lB ðxi ÞÞ 6 1 ð16Þ
then we have
ðt þ 1Þ 6 ðt v A ðxi Þ  lA ðxi ÞÞ  ðtv B ðxi Þ  lB ðxi ÞÞ 6 ðt þ 1Þ ð17Þ
It means that
0 6 jðtv A ðxi Þ  lA ðxi ÞÞ  ðt v B ðxi Þ  lB ðxi ÞÞjp 6 ðt þ 1Þp ð18Þ
Finally we have the following inequality:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 Xn
fjtðlA ðxi Þ  lB ðxi ÞÞ  ðv A ðxi Þ  v B ðxi ÞÞjp þ jtðv A ðxi Þ  v B ðxi ÞÞ  ðlA ðxi Þ  lB ðxi ÞÞjp g 6 1
p
06 p
2nðt þ 1Þ i¼1

Thus, 0 6 D(A,B) 6 1. h

A(2) Let A and B be two IFSs, if A = B then lA(xi) = lB(xi), vA(xi) = vB(xi) lA (xi)  lB(xi) = 0 and vA(xi)  vB(xi) = 0. Therefore,
the distance measure, D(A, B), is equal to zero. h

A(3) Let A and B be two IFSs. We can write the following equations:
jtðlA ðxi Þ  lB ðxi ÞÞ  ðv A ðxi Þ  v B ðxi ÞÞjp ¼ jð1ÞftðlB ðxi Þ  lA ðxi ÞÞ  ðv B ðxi Þ  v A ðxi ÞÞgjp ð19Þ
jtðv A ðxi Þ  v B ðxi ÞÞ  ðlA ðxi Þ  lB ðxi ÞÞjp ¼ jð1Þftðv B ðxi Þ  v A ðxi ÞÞ  ðlB ðxi Þ  lA ðxi ÞÞgjp ð20Þ
Based on the definition of absolute value, we have

jtðlA ðxi Þ  lB ðxi ÞÞ  ðv A ðxi Þ  v B ðxi ÞÞjp ¼ jtðlB ðxi Þ  lA ðxi ÞÞ  ðv B ðxi Þ  v A ðxi ÞÞjp ð21Þ
jtðv A ðxi Þ  v B ðxi ÞÞ  ðlA ðxi Þ  lB ðxi ÞÞjp ¼ jtðv B ðxi Þ  v A ðxi ÞÞ  ðlB ðxi Þ  lA ðxi ÞÞjp ð22Þ
Thus, D(A, B) = D(B, A). h

A(4) Let A, B, and C be three IFSs. The distance measures between A and B, and A and C are the followings:

DðA; BÞ ¼
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pn p p
i¼1 fjtðlA ðxi Þ  lB ðxi ÞÞ  ðv A ðxi Þ  v B ðxi ÞÞj þ jtðv A ðxi Þ  v B ðxi ÞÞ  ðlA ðxi Þ  lB ðxi ÞÞj g
p 1
2nðtþ1Þp

DðA; CÞ
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pn p p
i¼1 fjtðlA ðxi Þ  lC ðxi ÞÞ  ðv A ðxi Þ  v C ðxi ÞÞj þ jtðv A ðxi Þ  v C ðxi ÞÞ  ðlA ðxi Þ  lC ðxi ÞÞj g
1
¼ p 2nðtþ1Þ p

We can write the following equations:


jtðlA ðxi Þ  lC ðxi ÞÞ  ðv A ðxi Þ  v C ðxi ÞÞj ¼ jðt lA ðxi Þ  v A ðxi ÞÞ  ðt lC ðxi Þ  v C ðxi ÞÞj ð23Þ
jtðlA ðxi Þ  lB ðxi ÞÞ  ðv A ðxi Þ  v B ðxi ÞÞj ¼ jðt lA ðxi Þ  v A ðxi ÞÞ  ðt lB ðxi Þ  v B ðxi ÞÞj ð24Þ
jtðv A ðxi Þ  v B ðxi ÞÞ  ðlA ðxi Þ  lB ðxi ÞÞj ¼ jðt v A ðxi Þ  lA ðxi ÞÞ  ðt v B ðxi Þ  lB ðxi ÞÞj ð25Þ
jtðv A ðxi Þ  v C ðxi ÞÞ  ðlA ðxi Þ  lC ðxi ÞÞj ¼ jðt v A ðxi Þ  lA ðxi ÞÞ  ðt v C ðxi Þ  lC ðxi ÞÞj ð26Þ
If A # B # C, then lC(xi) P lB(xi) P lA(xi) and vA(xi) P vB(xi) P vC(xi). We also know from Definition 2 that 0 6 lA(xi) 6 lB(-
6 lA(xi) 6 lB(xi) 6 lC(xi) 6 1 and 0 6 vC(xi) 6 vB(xi) 6 vA(xi) 6 1.
Therefore, we have the following inequalities:
ðt lC ðxi Þ  v C ðxi ÞÞ P ðtlB ðxi Þ  v B ðxi ÞÞ P ðt lA ðxi Þ  v A ðxi ÞÞ ð27Þ
ðt v A ðxi Þ  lA ðxi ÞÞ P ðtv B ðxi Þ  lB ðxi ÞÞ P ðt v C ðxi Þ  lC ðxi ÞÞ ð28Þ
F.E. Boran, D. Akay / Information Sciences 255 (2014) 45–57 49

So it is easy to see that:

jðt lA ðxi Þ  v A ðxi ÞÞ  ðtlC ðxi Þ  v C ðxi ÞÞjp P jðt lA ðxi Þ  v A ðxi ÞÞ  ðt lB ðxi Þ  v B ðxi ÞÞjp ð29Þ
jðt v A ðxi Þ  lA ðxi ÞÞ  ðtv C ðxi Þ  lC ðxi ÞÞjp P jðt v A ðxi Þ  lA ðxi ÞÞ  ðt v B ðxi Þ  lB ðxi ÞÞjp ð30Þ
and finally we have the inequalities: D(A, C) P D(A, B) and D(A, C) P D(B, C).
We can say that D(A, B) is a distance measure between IFSs A and B sinceD(A, B) satisfies (A1)–(A4). h

Theorem 2. Dw(A, B) is a weighted distance measure between two IFSs A and B in X.

sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 Xn
w fjtðlA ðxi Þ  lB ðxi ÞÞ  ðv A ðxi Þ  v B ðxi ÞÞjp þ jtðv A ðxi Þ  v B ðxi ÞÞ  ðlA ðxi Þ  lB ðxi ÞÞjp g
p
Dw ðA; BÞ ¼ p i¼1 i
2ðt þ 1Þ
where t ¼ 2; 3; 4; . . . and p ¼ 1; 2; 3; . . .
ð31Þ
Pn
where wi is the weights of the features (xi)wi 2 [0, 1] and i¼1 wi ¼ 1.

Proof 2.

A(1) If we product the inequality defined in Eq. (14) with wi, then we have
0 6 wi jtðlA ðxi Þ  lB ðxi ÞÞ  ðv A ðxi Þ  v B ðxi ÞÞjp 6 wi ðt þ 1Þp ð32Þ

We can write the following inequality based on Eq. (32) as follows:

X
n X
n
06 wi jtðlA ðxi Þ  lB ðxi ÞÞ  ðv A ðxi Þ  v B ðxi ÞÞjp 6 wi ðt þ 1Þp ð33Þ
i¼1 i¼1
P P
It is easy to see that ni¼1 wi ðt þ 1Þp is equal to (t + 1)p in Eq. (33) since ni¼1 wi ¼ 1.
Similarly, if we product the inequality defined in Eq. (18) with wi, then we have

X
n X
n
06 wi jðtv A ðxi Þ  lA ðxi ÞÞ  ðt v B ðxi Þ  lB ðxi ÞÞjp 6 wi ðt þ 1Þp ð34Þ
i¼1 i¼1
P P
It is easy to see that ni¼1 wi ðt þ 1Þp is equal to (t + 1)p in Eq. (34) since ni¼1 wi ¼ 1.
Finally, we have the following inequality considering Eqs. (33) and (34):

sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 Xn
w fjtðlA ðxi Þ  lB ðxi ÞÞ  ðv A ðxi Þ  v B ðxi ÞÞjp þ jtðv A ðxi Þ  v B ðxi ÞÞ  ðlA ðxi Þ  lB ðxi ÞÞjp g 6 1
p
06 i¼1 i
2ðt þ 1Þp

Thus, 0 6 Dw(A,B) 6 1. h

A(2) Let A and B be two IFSs, if A = B then lA(xi) = lB(xi), vA(xi) = vB(xi) lA (xi)  lB(xi) = 0 and vA(xi)  vB(xi) = 0. Therefore,
the distance measure, Dw(A, B), is equal to zero. h

A(3) If we product both sides defined in Eqs. (21) and (22) with wi we have

wi jtðlA ðxi Þ  lB ðxi ÞÞ  ðv A ðxi Þ  v B ðxi ÞÞjp ¼ wi jtðlB ðxi Þ  lA ðxi ÞÞ  ðv B ðxi Þ  v A ðxi ÞÞjp ð35Þ
p p
wi jtðv A ðxi Þ  v B ðxi ÞÞ  ðlA ðxi Þ  lB ðxi ÞÞj ¼ wi jtðv B ðxi Þ  v A ðxi ÞÞ  ðlB ðxi Þ  lA ðxi ÞÞj ð36Þ
Thus,Dw(A,B) = Dw(B,A). h

A(4) If we product the inequalities defined in Eqs. (29) and (30) with wi, then we have

wi jðtlA ðxi Þ  v A ðxi ÞÞ  ðt lC ðxi Þ  v C ðxi ÞÞjp P wi jðt lA ðxi Þ  v A ðxi ÞÞ  ðt lB ðxi Þ  v B ðxi ÞÞjp ð37Þ
wi jðtv A ðxi Þ  lA ðxi ÞÞ  ðt v C ðxi Þ  lC ðxi ÞÞjp P wi jðt v A ðxi Þ  lA ðxi ÞÞ  ðt v B ðxi Þ  lB ðxi ÞÞjp ð38Þ
since all wi P 0. Finally we the following inequalities Dw (A, C) P Dw(A, B) and Dw(A, C) P Dw(B, C).

We can say that Dw(A, B) is a distance measure between IFSs A and B since Dw(A, B) satisfies (A1)–(A4). h
50 F.E. Boran, D. Akay / Information Sciences 255 (2014) 45–57

Theorem 3. If D(A, B) is a distance measure between IFSs A and B, then SpE ðA; BÞ ¼ 1  DðA; BÞ is a similarity measure between A
and B.
SpE ðA; BÞ ¼ 1
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Xn 1
fjtðlA ðxi Þ  lB ðxi ÞÞ  ðv A ðxi Þ  v B ðxi ÞÞjp þ jtðv A ðxi Þ  v B ðxi ÞÞ  ðlA ðxi Þ  lB ðxi ÞÞjp g
p
 i¼1 2nðt þ 1Þp

ð39Þ

Proof 3. From the axiom definitions of distance and similarity measures, it is easy to get the result. h

4. Interpretation of novel distance measure

In this section, we briefly interpret the novel distance measure and explain the functionality of parameter t defined in
distance measure.
Let (lA, vA) and (lB, vB) be two IFNs. lA can be equal to any value in [lA, lA + pA] and vA can be equal to any value in [vA, -
vA + pA]. The possible values for (lA, vA) are illustrated in Fig. 1 as the shaded area.
Undoubtedly, the most informative point among the all possible points in triangle ABC is D, the center of gravity of tri-
   
angle ABC. The pair l0A ; v 0A is easily found as lA þ p3A ; v A þ p3A . Based on the Definition 3, we can get the following equality
such that:
    
pA pA  3lA þ ð1  lA  v A Þ 3v A þ ð1  lA  v A Þ 1 þ 2lA  v A 1 þ 2v A  lA
lA þ ; vA þ ¼ ; ¼ ; ð40Þ
3 3 3 3 3 3
 
Similarly, we also have the pair l0B ; v B for IFN (lB,vB) as follows:
0

 
 0 0 1 þ 2lB  v B 1 þ 2v B  lB
lB ; v B ¼ ; ð41Þ
3 3
   
The absolute difference between l0A ; v 0A and l0B ; v 0B is defined in Eqs. (42) and (43):

l  l0 ¼ 1 þ 2lA  v A  1 þ 2lB  v B ¼ 2ðlA  lB Þ  ðv A  v B Þ
0
A B ð42Þ
3 3 3

0
v  v 0 ¼ 1 þ 2 v A  l A 1 þ 2 v B  l
B

2ðv A  v B Þ  ðl A  l Þ
B

A B  ¼ ð43Þ
3 3 3
0
l  l0 and v 0  v 0 to the power of p is equal to Eqs. (44) and (45) as follows:
A B A B

0
l  l0 p ¼ 1p j2ðl  l Þ  ðv A  v B Þjp ð44Þ
A B A B
3
0 p 1
v  v 0 ¼ j2ðv A  v B Þ  ðl  l Þjp ð45Þ
A B
3p A B
p p
The mean of l0A  l0B and v 0A  v 0B is obtained in Eq. (46) as follows:
1 n 0 o 1
lA  l0B p þ v 0A  v 0B p ¼ fj2ðlA  lB Þ  ðv A  v B Þjp þ j2ðv A  v B Þ  ðlA  lB Þjp g ð46Þ
2 2  3p

μA + π A
A

μ ′A D

μA
B C

v
vA v′A vA + π A

Fig. 1. The possible values for (lA, vA).


F.E. Boran, D. Akay / Information Sciences 255 (2014) 45–57 51

p p
The p root of the mean of l0A  l0B and v 0A  v 0B is found as:
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
p 1
n o 1

l0  l0 p þ v 0  v 0 p ¼ p j2ðlA  lB Þ  ðv A  v B Þjp þ j2ðv A  v B Þ  ðlA  lB Þjp ð47Þ
2 A B A B
2  3p
For more than one feature such as (xi),i = 1, 2, . . . , n, Eq. (47) can be defined in Eq. (48) as follows:
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Xn 1 n o
p
l0 ðxi Þ  l0 ðxi Þ p þ v 0 ðxi Þ  v 0 ðxi Þ p
i¼1 2n A B A B
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
p 1 Xn
¼ p fj2ðlA ðxi Þ  lB ðxi ÞÞ  ðv A ðxi Þ  v B ðxi ÞÞjp þ j2ðv A ðxi Þ  v B ðxi ÞÞ  ðlA ðxi Þ  lB ðxi ÞÞjp g ð48Þ
2n  3 i¼1
   
pA pA
It is easy to see that Eq. (48) is the special case of Eq. (8) when t = 2. If we parameterize l0A ; v A ¼ lA þ tþ1 ; v A þ tþ1 and 0
 0 0  pB pB

lB ; v B ¼ lB þ tþ1 ; v B þ tþ1 , we shall have Eq. (48) as the distance measure between (lA, vA) and (lB, vB). In Eq. (8), t is the
parameter adjusting the effect of hesitation margin in the computation. If t is very high, then the effect of hesitation margin is
neglected in the computation. If t is very low, then the effect of hesitation margin is taken into consideration in the
computation.

5. A comparison of similarity measures for IFSs

In order to illustrate the superiority of the proposed similarity measure, a comparison between the proposed similarity
measure and all the existing similarity measures is conducted. The existing similarity measures are defined in Table 1.
Table 2 presents a comprehensive comparison of the similarity measures for IFS with counter-intuitive examples (p = 1 in
SHB ; Spe ; Sps ; Sph and p = 1, t = 2 in SpE ). It is clearly seen that the second axiom of similarity measure (A6) is not satisfied by SC(A, B),
SDC(A, B), CIFS(A, B) since SC(A, B) = SDC(A, B) = CIFS(A, B)=1 when A = (0.3, 0.3) and B = (0.4, 0.4) which are indeed not equal to
each other. Similarly, the second axiom of similarity measure (A6) is also not satisfied by SC(A, B), SDC(A, B) when A =
(0.5, 0.5), B = (0, 0) and A = (0.4, 0.2), B = (0.5, 0.3). Some similarity measures have no capabilities to distinguish positive dif-
ference from negative difference. For example, SH(A, B) = SH(C, D) = 0.9 when A = (0.3, 0.3), B = (0.4, 0.4), C = (0.3, 0.4) and
D = (0.4, 0.3). The same counter-intuitive example exists for SO ; SHB ; Spe ; S1HY ; S2HY and S3HY . Another type of counter-intuitive case
occurs when A = (1, 0), B = (0, 0), C = (0.5, 0.5). In this situation, SH(A, B) and SH(C, B) are equal to 0.5. The same counter-intu-
itive example exists for SHB, Spe and CIFS. Another type of counter-intuitive example can be given for the case in which the
similarity measures are SL(A, B) = SL(A, C) = 0.95 when A = (0.4, 0.2), B = (0.5, 0.3), C = (0.5, 0.2). The same counter-intuitive
example also exists for Sps S1HY ; S2HY and S3HY . An interesting counter-intuitive case occurs when A = (0.4, 0.2), B = (0.5, 0.3),
C = (0.5, 0.2). In this case, it is expected that the similarity degree between A and B is equal or greater than the similarity de-
gree between A and C since they are ordered as C P B P A according to score function and accuracy function given in Def-
inition 3. However, the similarity degree between A and C is greater than the similarity degree between A and B when SH, SO,
SHB,Spe and Sph are used, which does not seem to be reasonable. On the other hand, SpE ðA; BÞ ¼ 0:967 and SpE ðA; CÞ ¼ 0:95. There-
fore, the proposed distance measure is in agreement with score function and accuracy function. The proposed distance mea-
sure is the only measure that have no the counter-intuitive cases as illustrated in Table 1 [14].

6. Numerical applications to pattern recognition

In order to illustrate the usefulness of the proposed distance measure for IFS to pattern recognition, we present some
examples in this section.

Example 1. Three known patterns, P1, P2, and P3, with class labels C1, C2, and C3, respectively, are given. The patterns are
represented by the following IFSs in X = {x1, x2, x3, x4}:

P1 ¼ fhx1 ; 0:5; 0:3jx1 2 Xi; hx2 ; 0:7; 0:0jx2 2 Xi; hx3 ; 0:4; 0:5jx3 2 Xi; hx4 ; 0:7; 0:3jx4 2 Xig
P2 ¼ fhx1 ; 0:5; 0:2jx1 2 Xi; hx2 ; 0:6; 0:1jx2 2 Xi; hx3 ; 0:2; 0:7jx3 2 Xi; hx4 ; 0:7; 0:3jx4 2 Xig
P3 ¼ fhx1 ; 0:5; 0:4jx1 2 Xi; hx2 ; 0:7; 0:1jx2 2 Xi; hx3 ; 0:4; 0:6jx3 2 Xi; hx4 ; 0:7; 0:2jx4 2 Xig
Unknown pattern Q is given as follows:

Q ¼ fhx1 ; 0:4; 0:3jx1 2 Xi; hx2 ; 0:7; 0:1jx2 2 Xi; hx3 ; 0:3; 0:6jx3 2 Xi; hx4 ; 0:7; 0:3jx4 2 Xig
It is aimed to find out the class that Q belongs to. In order to do that, the similarity degrees between Q and classes C1, C2 and
C3 are calculated, and Q is then assigned to C k described by:

k ¼ arg maxfSðPk ; Q Þg ð49Þ
k

The similarity degree between S(P1,Q), S(P2, Q) and S(P3, Q) are calculated for all existing similarity measures and shown in
Table 3 as follows:
52 F.E. Boran, D. Akay / Information Sciences 255 (2014) 45–57

Table 1
Existing similarity measures.

Authors Similarity measure


Pn
Chen [31] jðl ðxi Þv A ðxi ÞÞjlB ðxi Þv B ðxi Þjj
SC ðA; BÞ ¼ 1  i¼1 A 2n
Pn
Hong and Kim [32] jðl ðxi ÞlB ðxi ÞÞjv A ðxi Þv B ðxi Þjj
SH ðA; BÞ ¼ 1  i¼1 A 2n
Pn
Fan and Zhangyan [33] fjðl ðxi Þv A ðxi ÞÞðlB ðxi Þv B ðxi ÞÞjþjlA ðxi ÞlB ðxi Þjþjðv A ðxi Þv B ðxi ÞÞjg
SL ðA; BÞ ¼ 1  i¼1 A 4n
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pn
Li et al. [34] ðlA ðxi ÞlB ðxi ÞÞ2 þðv A ðxi Þv B ðxi ÞÞ2
SO ðA; BÞ ¼ 1  i¼1
2n
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pn ffi
Li and Cheng [11] p jw ðx ÞwB ðxi Þjp
i¼1 A i
SDC ðA; BÞ ¼ 1  n
wA ðxi Þ ¼ ðlA ðxi Þ þ 1  v A ðxi ÞÞ=2
wB ðxi Þ ¼ ðlB ðxi Þ þ 1  v B ðxi ÞÞ=2

Mitchell [12]

1
SHB ðA; BÞ ¼ ðql ðA; BÞ þ qv ðA; BÞÞ
2 s ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pn p Pn p
i¼1 jlA ðxi Þ  lB ðxi Þj i¼1 jv A ðxi Þ  v B ðxi Þj
p p
ql ðA; BÞ ¼ 1  qv ðA; BÞ ¼ 1 
n n

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pn ffi
Liang and Shi [13] p j/l ðxi Þþ/v ðxi Þj
p

Spe ðA; BÞ ¼ 1  i¼1


n
/l ðxi Þ ¼ jlA ðxi Þ  lB ðxi Þj=2
/v ðxi Þ ¼ jð1  v A ðxi ÞÞ  ð1  v B ðxi ÞÞj=2

Liang and Shi [13]


sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pn p
i¼1 jus1 ðxi Þ þ us2 ðxi Þj
p
Sps ðA; BÞ ¼1
n
us1 ðxi Þ ¼ jmA1 ðxi Þ  mB1 ðxi Þj=2;
us1 ðxi Þ ¼ jmA2 ðxi Þ  mB2 ðxi Þj=2
mA1 ðxi Þ ¼ jlA ðxi Þ þ mA ðxi Þj=2;
mB1 ðxi Þ ¼ jlB ðxi Þ þ mB ðxi Þj=2;
mA2 ðxi Þ ¼ j1  v A ðxi Þ þ mA ðxi Þj=2;
mB1 ðxi Þ ¼ j1  v B ðxi Þ þ mB ðxi Þj=2;
mA ðxi Þ ¼ jlA ðxi Þ þ 1  v A ðxi Þj=2;
mB ðxi Þ ¼ jlB ðxi Þ þ 1  v B ðxi Þj=2

Liang and Shi [13]


 
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi g1 ðiÞ ¼ /l ðxi Þ þ /v ðxi Þ Spe
Pn p
p p
i¼1 ðg1 ðiÞ þ g2 ðiÞ þ g3 ðiÞÞ
g2 ðiÞ ¼ jwA ðxi Þ  wB ðxi ÞjðSDC Þ
Sh ðA; BÞ ¼ 1 
3n g3 ðiÞ ¼ maxðlA ðiÞ; lB ðiÞÞ  minðlA ðiÞ; lB ðiÞÞ
 
lA ðiÞ ¼ 1  lA ðxi Þ  v A ðxi Þ =2 lB ðiÞ ¼ ð1  lB ðxi Þ  v B ðxi ÞÞ=2

Hung and Yang [10]

edH ðA;BÞ  e1 3 1  dH ðA; BÞ


S1HY ðA; BÞ ¼ 1  dH ðA; BÞ S2HY ðA; BÞ ¼ SHY ðA; BÞ ¼
1  e1 1 þ dH ðA; BÞ
1X n
dH ðA; BÞ ¼ maxðjlA ðxi Þ  lB ðxi Þj; jv A ðxi Þ  v B ðxi ÞjÞ
n i¼1

Ye [15] 1 Xn lA ðxi ÞlB ðxi Þ þ v A ðxi Þv B ðxi Þ


C IFS ðA; BÞ ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
n i¼1
l2 ðx Þ þ v 2 ðx Þ l2 ðx Þ þ v 2 ðx Þ
A i A i B i B i
F.E. Boran, D. Akay / Information Sciences 255 (2014) 45–57 53

Table 2
The comparison of similarity measures (counter-intuitive cases are in bold type). (p = 1 in SHB ; Spe ; Sps ; Sph and p = 1, t = 2 in SpE ).

1 2 3 4 5 6
A (0.3, 0.3) (0.3, 0.4) (1, 0) (0.5, 0.5) (0.4, 0.2) (0.4, 0.2)
B (0.4, 0.4) (0.4, 0.3) (0, 0) (0, 0) (0.5, 0.3) (0.5, 0.2)
SC 1 0.9 0.5 1 1 0.95
SH 0.9 0.9 0.5 0.5 0.9 0.95
SL 0.95 0.9 0.5 0.75 0.95 0.95
SO 0.9 0.9 0.3 0.5 0.9 0.93
SDC 1 0.9 0.5 1 1 0.95
SHB 0.9 0.9 0.5 0.5 0.9 0.95
Spe 0.9 0.9 0.5 0.5 0.9 0.95
Sps 0.95 0.9 0.5 0.75 0.95 0.95
Sph 0.93 0.933 0.5 0.67 0.93 0.95
S1HY 0.9 0.9 0 0.5 0.9 0.9
S2HY 0.85 0.85 0 0.38 0.85 0.85
S3HY 0.82 0.82 0 0.33 0.82 0.82
CIFS 1 0.96 0 0 0.9971 0.9965
SpE 0.967 0.9 0.5 0.833 0.967 0.95

Table 3
The similarity measures between the known patterns and the unknown pattern in Example 1 (Patterns not discriminated are in bold type.) (p = 1 in SHB ; Spe ; Sps ; Sph
and p = 1, t = 2 in SpE ).

S(P1, Q) S(P2, Q) S(P3, Q) S(P1, Q) S(P2, Q) S(P3, Q)


SC 0.925 0.863 0.925 Sps 0.950 0.938 0.963
SH 0.975 0.963 0.975 Sph 0.958 0.954 0.958
SL 0.950 0.938 0.963 S1HY 0.925 0.925 0.925
SO 0.929 0.921 0.929 S2HY 0.886 0.886 0.886
SDC 0.950 0.938 0.975 S3HY 0.860 0.860 0.860
SHB 0.950 0.938 0.950 CIFS 0.991 0.987 0.996
Spe 0.950 0.938 0.950 SpE 0.950 0.938 0.967

The results obtained in Table 3 indicate that S1HY ; S2HY and S3HY cannot capable of discriminating the difference between the
known patterns P1,P2 and P3 and unknown pattern Q since the similarity degrees between known patterns and unknown
pattern are equal to each other. In some of the other cases in which the similarity degrees between known patterns P1, P3
and unknown pattern Q are equal to each other when SC, SH,SO, SHB, Sph and Spe are used. The difference between the patterns
is discriminated when SL, SDC, Sps ,CIFS and SpE are used as the similarity measure and unknown pattern Q belongs to C3. How-
ever, this does not mean that SC, SL, SDC, Sps ,CIFS and SpE are capable of differentiating patterns. The following two examples
illustrate some special situations in pattern recognition.

Example 2. Three known patterns, P1, P2, and P3, with class labels C1, C2, and C3, respectively, are given. The patterns are
represented by the following IFSs in X = {x1, x2, x3, x4}:
P1 ¼ fhx1 ; 0:3; 0:3jx1 2 Xi; hx2 ; 0:4; 0:4jx2 2 Xi; hx3 ; 0:4; 0:4jx3 2 Xi; hx4 ; 0:4; 0:4jx4 2 Xig
P2 ¼ fhx1 ; 0:5; 0:5jx1 2 Xi; hx2 ; 0:1; 0:1jx2 2 Xi; hx3 ; 0:5; 0:5jx3 2 Xi; hx4 ; 0:1; 0:1jx4 2 Xig
P3 ¼ fhx1 ; 1:0; 0:0jx1 2 Xi; hx2 ; 0:3; 0:3jx2 2 Xi; hx3 ; 0:5; 0:5jx3 2 Xi; hx4 ; 0:6; 0:0jx4 2 Xig
Unknown pattern Q is given as follows:

Q ¼ fhx1 ; 0:4; 0:4jx1 2 Xi; hx2 ; 0:5; 0:5jx2 2 Xi; hx3 ; 0:2; 0:2jx3 2 Xi; hx4 ; 0:3; 0:3jx4 2 Xig

Table 4
The similarity measures between the known patterns and the unknown pattern in Example 2 (Patterns not discriminated are in bold type.) (p = 1 in SHB ; Spe ; Sps ; Sph
and p = 1, t = 2 in SpE ).

S(P1, Q) S(P2, Q) S(P3, Q) S(P1, Q) S(P2, Q) S(P3, Q)


SC 1.000 1.000 0.800 Sps 0.938 0.875 0.738
SH 0.950 0.850 0.925 Sph 0.917 0.833 0.775
SL 0.938 0.875 0.738 S1HY 0.875 0.750 0.650
SO 0.868 0.726 0.654 S2HY 0.814 0.650 0.533
SDC 1.000 1.000 0.800 S3HY 0.778 0.600 0.481
SHB 0.875 0.750 0.675 CIFS 1.000 1.000 0.854
Spe 0.875 0.750 0.675 SpE 0.958 0.917 0.758
54 F.E. Boran, D. Akay / Information Sciences 255 (2014) 45–57

It is aimed to find out the class that Q belongs to. In order to do that, the similarity degrees between Q and classes C1, C2 and
C3 are calculated, and Q is then assigned to C k described by:

k ¼ rmargmaxk fSðP k ; Q Þg ð50Þ
The similarity degree between S(P1, Q), S(P2, Q) and S(P3, Q) are calculated for all existing similarity measures and shown in
Table 4 as follows:
The results obtained in Table 4 indicate that SC, SDC, and CIFS are not capable of discriminating the difference between the
known patterns P1, P2 and unknown pattern Q since the similarity degrees between known patterns and unknown pattern
are equal to one even though patterns are not identical. Therefore, unknown pattern cannot be properly classified by using
SC, SDC and CIFS. Moreover, they do not satisfy the second axiom of similarity measure (A6). SH, SL, SO, SHB, Spe ; Sph ; Sps S1HY ; S2HY ; S3HY
and SpE can capable of discriminating the difference, and assign unknown pattern Q to C1.

Example 3. Three known patterns, P1, P2, and P3, with class labels C1, C2, and C3, respectively, are given. The patterns are
represented by the following IFSs in X = {x1, x2, x3, x4}:

P1 ¼ fhx1 ; 0:5; 0:2jx1 2 Xi; hx2 ; 0:5; 0:2jx2 2 Xi; hx3 ; 0:4; 0:2jx3 2 Xi; hx4 ; 0:5; 0:3jx4 2 Xig
P2 ¼ fhx1 ; 0:5; 0:3jx1 2 Xi; hx2 ; 0:5; 0:2jx2 2 Xi; hx3 ; 0:4; 0:2jx3 2 Xi; hx4 ; 0:3; 0:5jx4 2 Xig
P3 ¼ fhx1 ; 0:3; 0:1jx1 2 Xi; hx2 ; 0:5; 0:0jx2 2 Xi; hx3 ; 0:3; 0:1jx3 2 Xi; hx4 ; 0:5; 0:5jx4 2 Xig
Unknown pattern Q is given as follows:

Q ¼ fhx1 ; 0:4; 0:2jx1 2 Xi; hx2 ; 0:5; 0:2jx2 2 Xi; hx3 ; 0:4; 0:2jx3 2 Xi; hx4 ; 0:5; 0:5jx4 2 Xig
It is aimed to find out the class that Q belongs to. In order to do that, the similarity degrees between Q and classes C1, C2 and
C3 are calculated, and Q is then assigned to C k described by:

k ¼ argmaxk fSðPk ; Q Þg ð51Þ
The similarity degree between S(P1, Q), S(P2, Q) and S(P3, Q) are calculated for all existing similarity measures and shown in
Table 5 as follows:
As it is clearly seen in Table 5, the similarity degrees S(P1, Q) and S(P2, Q) are equal to each other when SL,Sps ; S1HY ; S2HY and S3HY
are used. Furthermore, the similarity degrees S(P2, Q)S(P3, Q) are equal to each other when SC andSDC are used. These two cases
indicate that these similarity measures cannot capable of discriminating difference between patterns. Thus unknown pattern
cannot be classified properly when these measure are used. On the other hand, SO, SHB, Spe ; Sph , CIFS and SpE can capable of
discriminating the difference. However, unknown pattern Q belongs to class C1 when SO, SHB, Spe and Sph are used; unknown

Table 5
The similarity measures between the known patterns and the unknown pattern in Example 3 (Patterns not discriminated are in bold type.) (p = 1 in SHB ; Spe ; Sps ; Sph
and p = 1, t = 2 in SpE ).

S(P1, Q) S(P2, Q) S(P3, Q) S(P1, Q) S(P2, Q) S(P3, Q)


SC 0.963 0.975 0.975 Sps 0.963 0.963 0.950
SH 0.963 0.975 0.925 Sph 0.963 0.958 0.942
SL 0.963 0.963 0.950 S1HY 0.925 0.925 0.900
SO 0.921 0.913 0.900 S2HY 0.886 0.886 0.849
SDC 0.963 0.975 0.975 S3HY 0.860 0.860 0.818
SHB 0.963 0.950 0.925 CIFS 0.9917 0.9918 0.977
Spe 0.963 0.950 0.925 SpE 0.963 0.967 0.958

Table 6
Four classes of known building materials and unknown building material represented by IFNs (p = 1, t = 2 in SpE ).

x1 x2 x3 x4 x5 x6 x7 x8 x9 x10 x11 x12


lA1 ðxÞ 0.173 0.102 0.53 0.965 0.42 0.008 0.331 1 0.215 0.432 0.75 0.432
vA ðxÞ
1
0.524 0.818 0.326 0.008 0.351 0.956 0.512 0 0.625 0.534 0.126 0.432
lA2 ðxÞ 0.51 0.627 1 0.125 0.026 0.732 0.556 0.65 1 0.145 0.047 0.76
vA 2
ðxÞ 0.365 0.125 0 0.648 0.823 0.153 0.303 0.267 0 0.762 0.923 0.231
lA3 ðxÞ 0.495 0.603 0.987 0.073 0.037 0.69 0.147 0.213 0.501 1 0.324 0.045
vA 3
ðxÞ 0.387 0.298 0.006 0.849 0.923 0.268 0.812 0.653 0.284 0 0.483 0.912
lA4 ðxÞ 1 1 0.857 0.734 0.021 0.076 0.152 0.113 0.489 1 0.386 0.028
vA 5
ðxÞ 0 0 0.123 0.158 0.896 0.912 0.712 0.756 0.389 0 0.485 0.912
lB(x) 0.978 0.98 0.798 0.693 0.051 0.123 0.152 0.113 0.494 0.987 0.376 0.012
vB(x) 0.003 0.012 0.132 0.213 0.876 0.756 0.721 0.732 0.368 0 0.423 0.897
F.E. Boran, D. Akay / Information Sciences 255 (2014) 45–57 55

Table 7
Symptoms characteristic for the patients.

Temperature Headache Stomach pain Cough Chest pain


Al (0.8, 0.1) (0.6, 0.1) (0.2, 0.8) (0.6, 0.1) (0.1, 0.6)
Bob (0.0, 0.8) (0.4, 0.4) (0.6, 0.1) (0.1, 0.7) (0.1, 0.8)
Joe (0.8, 0.1) (0.8, 0.1) (0.0, 0.6) (0.2, 0.7) (0.0, 0.5)
Ted (0.6, 0.1) (0.5, 0.4) (0.3, 0.4) (0.7, 0.2) (0.3, 0.4)

Table 8
Symptoms characteristic for the diagnoses.

Temperature Headache Stomach pain Cough Chest pain


Viral Fever (0.4, 0.0) (0.3, 0.5) (0.1, 0.7) (0.4, 0.3) (0.1, 0.7)
Malaria (0.7, 0.0) (0.2, 0.6) (0.0, 0.9) (0.7, 0.0) (0.1, 0.8)
Typhoid (0.3, 0.3) (0.6, 0.1) (0.2, 0.7) (0.2, 0.6) (0.1, 0.9)
Stomach problem (0.1, 0.7) (0.2, 0.4) (0.8, 0.0) (0.2, 0.7) (0.2, 0.7)
Chest problem (0.1, 0.8) (0.0, 0.8) (0.2, 0.8) (0.2, 0.8) (0.8, 0.1)

Table 9
The results measured by the proposed similarity measure.

Viral fever Malaria Typhoid Stomach problem Chest problem


Al 0.843 0.833 0.800 0.553 0.500
Bob 0.670 0.540 0.740 0.900 0.653
Joe 0.803 0.713 0.860 0.627 0.567
Ted 0.820 0.770 0.710 0.630 0.550

pattern Q belongs to class C2 when SH, CIFS and SpE are used. It is easy to see that the reason for this difference is the first fea-
ture, i.e., (x1). The IFNs of x1 are (0.5, 0.2), (0.5, 0.3) and (0.4, 0.2) for P1, P2 and Q, respectively. It is expected that the similarity
degree between (0.5, 0.3) and (0.4, 0.2) is at least equal or greater than the similarity degree between (0.5,0.2) and (0.4,0.2)
since they are ordered as (0.5, 0.2) > (0.5, 0.3) > (0.4, 0.2) according to Definition 3. Therefore, it seems that S(P2, Q) > S(P1, Q) is
more reasonable.

Example 4. [7] A1, A2, A3, A4, four classes of known building materials, and B, unknown building material, are given in the
feature space X = {x1, x2, . . . , x12} and represented by IFSs (Table 6). It is aimed to find out which class B belongs to.

In order to do that, the distances between four classes of building material and unknown building material are calculated
with (8) as follows:

DðA1 ; BÞ ¼ 0:431 DðA2 ; BÞ ¼ 0:434 DðA3 ; BÞ ¼ 0:197 DðA4 ; BÞ ¼ 0:025


SðA1 ; BÞ ¼ 0:569 SðA2 ; BÞ ¼ 0:566 SðA3 ; BÞ ¼ 0:803 SðA4 ; BÞ ¼ 0:975
B has been classified to A4, this result is the same with the result obtained in [7].

Example 5. [21,26–30]Let there be four patients Al, Bob, Joe, Ted, and the set of patients is represented as T = {Al, Bob, Joe,
Ted}. Their symptoms are Temperature, Headache, Stomach pain, Cough, Chest pain, and the set of symptoms is represented
as S={Temperature, Headache, Stomach pain, Cough, Chest pain}. The set of diagnosis is defined, i.e. D = {Viral Fever, Malaria,
Typhoid, Stomach problem, Heart problem}. IFNs are given as relation T ? S and S ? D in Tables 7 and 8.
In order to accomplish a proper diagnosis for each patient, the distance between a diagnosis and all patients is calculated
in context of symptoms. This process is done for all diagnoses. The higher similarity (the lower distance), pointing to a proper
diagnosis is intended. In Table 9, the degree of similarity between patients and diagnoses are presented. According to the
similarity degrees in Table 9, Al suffers from Viral Fever, Bob suffers from Stomach problem, Joe suffers from Typhoid,
and Ted suffers from Viral Fever. When the results are deeply analyzed, the similarity degrees of Viral Fever and Malaria
for Al are so closed. The same problem has been considered in [21,26–30]. The results obtained in these studies are shown
in Table 10. It is obviously seen that Bob suffers from a stomach problem since all the measures provide the same result. Joe
suffers from typhoid in seven out of the nine methods. Five out of the nine methods indicate that Al suffers from Viral Fever,
and other methods indicate that Al suffers from Malaria. Similarly, six out of the nine methods indicate that Ted suffers from
Viral Fever, and other methods indicate that Ted suffers from Malaria. This situation illustrates that it is very hard to
56 F.E. Boran, D. Akay / Information Sciences 255 (2014) 45–57

Table 10
All the considered results.

p = 0 in [28] p = 1 in [28] Szmidt and Kacprzyk [27]


Al Viral fever Viral fever Malaria
Bob Stomach problem Stomach problem Stomach problem
Joe Typhoid Stomach problem Typhoid
Ted Viral fever Viral fever Viral Fever
Szmidt and Kacprzyk in [26] De et al. in [29] Fuzzy similarity in [28]
Al Viral fever Malaria Malaria
Bob Stomach problem Stomach problem Stomach problem
Joe Typhoid Malaria Typhoid
Ted Malaria Malaria Malaria
Vlachos and Sergiadis [30] p = 1,t = 2 SpE Wei et al. in [21]
Al Viral fever Viral fever Malaria
Bob Stomach problem Stomach problem Stomach problem
Joe Typhoid Typhoid Typhoid
Ted Viral fever Viral fever Viral fever

diagnose whether Al and Ted suffers from Viral Fever or Malaria or not because these two symptoms are involved with each
other. The proposed method provides the same results obtained in Own [28] and Vlachos and Sergiadis [30] with less com-
putational burden.

7. Conclusion

Even though several similarity measures are proposed for IFSs in the literature, most of them have provided counter-intu-
itive results. In this study, we have proposed a new general type of similarity measure between IFSs depending on two
parameters, expressing Lp norm and the level of uncertainty, respectively. In some special cases, it has been seen that some
of the existing similarity measure cannot provide reasonable results. The proposed similarity measure is however capable of
differentiating IFSs in these special cases. Moreover, we have reinvestigated five problems encountered in pattern recogni-
tion by using the proposed method. For pattern recognition problems in Examples 1–4, the results are the same with some of
those of other studies. For Example 5, a compatible result is achieved compared to the results of previous studies. In the light
of this study, the proposed similarity measure can be effectively used in the real applications of decision making, pattern
recognition, linguistic summarization and data mining in the future research.

Acknowledgements

The authors are very grateful to the Editor-in-Chief, Professor Witold Pedrycz, and the anonymous referees, for their con-
structive comments and suggestions that led to an improved version of this paper. This research is supported by The Scien-
tific and Technological Research Council of Turkey (TUBITAK: 112M030).

References

[1] L.A. Zadeh, Fuzzy sets, Information and Control 8 (1965) 338–353.
[2] K.T. Atanassov, Intuitionistic Fuzzy-Sets, Fuzzy Sets and Systems 20 (1986) 87–96.
[3] W.L. Gau, D.J. Buehrer, Vague Sets, Ieee Transactions on Systems Man and Cybernetics 23 (1993) 610–614.
[4] H. Bustince, P. Burillo, Vague sets are intuitionistic fuzzy sets, Fuzzy Sets and Systems 79 (1996) 403–405.
[5] M.M. Xia, Z.S. Xu, Some new similarity measures for intuitionistic fuzzy values and their application in group decision making, Journal of Systems
Science and Systems Engineering 19 (2010) 430–452.
[6] E. Szmidt, J. Kacprzyk, Distances between intuitionistic fuzzy sets, Fuzzy Sets and Systems 114 (2000) 505–518.
[7] W.Q. Wang, X.L. Xin, Distance measure between intuitionistic fuzzy sets, Pattern Recognition Letters 26 (2005) 2063–2069.
[8] P. Grzegorzewski, Distances between intuitionistic fuzzy sets and/or interval-valued fuzzy sets based on the Hausdorff metric, Fuzzy Sets and Systems
148 (2004) 319–328.
[9] T.-Y. Chen, A note on distances between intuitionistic fuzzy sets and/or interval-valued fuzzy sets based on the Hausdorff metric, Fuzzy Sets and
Systems 158 (2007) 2523–2525.
[10] W.L. Hung, M.S. Yang, Similarity measures of intuitionistic fuzzy sets based on Hausdorff distance, Pattern Recognition Letters 25 (2004) 1603–1611.
[11] D.F. Li, C.T. Cheng, New similarity measures of intuitionistic fuzzy sets and application to pattern recognitions, Pattern Recognition Letters 23 (2002)
221–225.
[12] H.B. Mitchell, On the Dengfeng–Chuntian similarity measure and its application to pattern recognition, Pattern Recognition Letters 24 (2003) 3101–
3104.
[13] Z.Z. Liang, P.F. Shi, Similarity measures on intuitionistic fuzzy sets, Pattern Recognition Letters 24 (2003) 2687–2693.
[14] Y.H. Li, D.L. Olson, Z. Qin, Similarity measures between intuitionistic fuzzy (vague) sets: a comparative analysis, Pattern Recognition Letters 28 (2007)
278–285.
[15] J. Ye, Cosine similarity measures for intuitionistic fuzzy sets and their applications, Mathematical and Computer Modelling 53 (2011) 91–97.
[16] C.M. Hwang, M.S. Yang, W.L. Hung, M.G. Lee, A similarity measure of intuitionistic fuzzy sets based on the Sugeno integral with its application to
pattern recognition, Information Sciences 189 (2012) 93–109.
[17] Z.S. Xu, Some similarity measures of intuitionistic fuzzy sets and their applications to multiple attribute decision making, Fuzzy Optim Decis Ma 6
(2007) 109–121.
F.E. Boran, D. Akay / Information Sciences 255 (2014) 45–57 57

[18] Z.S. Xu, J. Chen, An overview of distance and similarity measures of intuitionistic fuzzy sets, Int J Uncertain Fuzz 16 (2008) 529–555.
[19] Z.S. Xu, R.R. Yager, Intuitionistic and interval-valued intutionistic fuzzy preference relations and their measures of similarity for the evaluation of
agreement within a group, Fuzzy Optim Decis Ma 8 (2009) 123–139.
[20] W.Y. Zeng, P. Guo, Normalized distance, similarity measure, inclusion measure and entropy of interval-valued fuzzy sets and their relationship,
Information Sciences 178 (2008) 1334–1342.
[21] C.P. Wei, P. Wang, Y.Z. Zhang, Entropy, similarity measure of interval-valued intuitionistic fuzzy sets and their applications, Information Sciences 181
(2011) 4273–4286.
[22] M.H. Shu, C.H. Cheng, J.R. Chang, Using intuitionistic fuzzy sets for fault-tree analysis on printed circuit board assembly, Microelectronics Reliability 46
(2006) 2139–2148.
[23] S.M. Chen, J.M. Tan, Handling multicriteria fuzzy decision-making problems based on vague set-theory, Fuzzy Sets and Systems 67 (1994) 163–172.
[24] D.H. Hong, C.H. Choi, Multicriteria fuzzy decision-making problems based on vague set theory, Fuzzy Sets and Systems 114 (2000) 103–113.
[25] Z.S. Xu, R.R. Yager, Some geometric aggregation operators based on intuitionistic fuzzy sets, International Journal of General Systems 35 (2006) 417–
433.
[26] E. Szmidt, J. Kacprzyk, A similarity measure for intuitionistic fuzzy sets and its application in supporting medical diagnostic reasoning, Artificial
Intelligence and Soft Computing – ICAISC 2004 3070 (2004) 388–393.
[27] E. Szmidt, J. Kacprzyk, Intuitionistic fuzzy sets in intelligent data analysis for medical diagnosis, Computational Science – ICCS 2001, Proceedings Pt 2
2074 (2001) 263–271.
[28] C.M. Own, Switching between type-2 fuzzy sets and intuitionistic fuzzy sets: an application in medical diagnosis, Applied Intelligence 31 (2009) 283–
291.
[29] S.K. De, R. Biswas, A.R. Roy, An application of intuitionistic fuzzy sets in medical diagnosis, Fuzzy Sets and Systems 117 (2001) 209–213.
[30] I.K. Vlachos, G.D. Sergiadis, Intuitionistic fuzzy information – applications to pattern recognition, Pattern Recognition Letters 28 (2007) 197–206.
[31] S.M. Chen, Measures of similarity between vague sets, Fuzzy Sets and Systems 74 (1995) 217–223.
[32] D.H. Hong, C. Kim, A note on similarity measures between vague sets and between elements, Information Sciences 115 (1999) 83–96.
[33] L. Fan, X. Zhangyan, Similarity measures between vague sets, J. Software 12 (2001) 922–927.
[34] Y. Li, C. Zhongxian, Y. Degin, Similarity measures between vague sets and vague entropy, Journal of Computer Science 29 (2002) 129–132.