A Comparative Study of Two-Sample t-Test Under Fuzzy Environments Using Trapezoidal Fuzzy Numbers

International Journal of Mathematics and Statistics Invention (IJMSI)
E-ISSN: 2321 – 4767 P-ISSN: 2321 - 4759
www.ijmsi.org Volume 4 Issue 1 || January. 2016 || PP-39-54
www.ijmsi.org 39 | Page
A Comparative Study of Two-Sample t-Test Under Fuzzy
Environments Using Trapezoidal Fuzzy Numbers
S. Parthiban1
and P. Gajivaradhan2
1
Research Scholar, Department of Mathematics, Pachaiyappa’s College,
Chennai-600 030, Tamil Nadu, India.
2
Department of Mathematics, Pachaiyappa’s College, Chennai-600 030, Tamil Nadu, India.
ABSTRACT: This paper proposes a method for testing hypotheses over two sample t-test under fuzzy
environments using trapezoidal fuzzy numbers (tfns.). In fact, trapezoidal fuzzy numbers have many advantages
over triangular fuzzy numbers as they have more generalized form. Here, we have approached a new method
where trapezoidal fuzzy numbers are defined in terms of alpha level of trapezoidal interval data and based on
this approach, the test of hypothesis is performed. Moreover the proposed test is analysed under various types
of trapezoidal fuzzy models such as Alpha Cut Interval, Membership Function, Ranking Function, Total Integral
Value and Graded Mean Integration Representation. And two numerical examples have been illustrated.
Finally a comparative view of all conclusions obtained from various test is given for a concrete comparative
study.
KEYWORDS: Trapezoidal Fuzzy Numbers (tfns./TFNS.), Alpha Cut, Test of Hypothesis, Confidence Limits,
Two-sample t-Test, Ranking Function, Total Integral Value (TIV), Graded Mean Integration Representation
(GMIR).
AMS Mathematics Subject Classification (2010): 62A86, 62F03, 97K80
I. INTRODUCTION
Statistical hypothesis testing is very important tool for finding decisions in practical problems.
Usually, the underlying data are assumed to be precise numbers but it is much more realistic in general to
consider fuzzy values which are non-precise numbers. In this case the test statistic will also yield a non-precise
number. This paper presents an approach for statistical hypothesis testing on the basis of alpha cuts of
(trapezoidal) fuzzy numbers. The statistical hypotheses testing under fuzzy environments has been studied by
many authors using the fuzzy set theory concepts introduced by Zadeh [40]. Viertl [35] investigated some
methods to construct confidence intervals and statistical tests for fuzzy data. Wu [39] proposed some
approaches to construct fuzzy confidence intervals for the unknown fuzzy parameter. A new approach to the
problem of testing statistical hypotheses is introduced by Chachi et al. [15]. Asady [9] introduced a method to
obtain the nearest trapezoidal approximation of fuzzy numbers. Gajivaradhan and Parthiban analysed one
sample t-test and two sample t-test using alpha cut interval method using trapezoidal fuzzy numbers [18, 19].
Abhinav Bansal [5] explored some arithmetic properties of arbitrary trapezoidal fuzzy numbers of the form
(a, b, c, d). Moreover, Liou and Wang ranked fuzzy numbers with total integral value [25]. Wang et al.
presented the method for centroid formulae for a generalized fuzzy number [37]. Iuliana Carmen
BĂRBĂCIORU dealt with the statistical hypotheses testing using membership function of fuzzy numbers [21].
Salim Rezvani analysed the ranking functions with trapezoidal fuzzy numbers [30]. Wang arrived some
different approach for ranking trapezoidal fuzzy numbers [37]. Thorani et al. approached the ranking function
of a trapezoidal fuzzy number with some modifications [31]. Salim Rezvani and Mohammad Molani presented
the shape function and Graded Mean Integration Representation (GMIR) for trapezoidal fuzzy numbers [29].
Liou and Wang proposed the Total Integral Value (TIV) of the trapezoidal fuzzy number with the index of
optimism and pessimism [25].
In this paper, we propose a new statistical fuzzy hypothesis testing of two-sample t-test in which the
designated samples are in terms of fuzzy (trapezoidal fuzzy numbers) data. Another idea in this paper is, when
we have some vague data about an experiment, what can be the result when the centroid point/ranking grades of
those imprecise data are employed in hypothesis testing? For this reason, we have used the centroid/ranking
grades of trapezoidal fuzzy numbers (tfns.) in hypothesis testing. In the decision rules of the proposed testing
technique, degrees of optimism, pessimism and h-level sets are not used. In fact, we would like to counter an
argument that the α-cut interval method can be general enough to deal with two-sample t-test under fuzzy
environments. And for better understanding, the proposed fuzzy hypothesis testing technique is illustrated with
two numerical examples at each models. Finally a tabular form of all conclusions obtained from various test is
given for a concrete comparative study. And the same concept can also be used when we have samples in terms
of triangular fuzzy numbers [10].

A Comparative Study of Two-Sample t-Test under…
II. PRELIMINARIES
Definition 2.1. Generalized fuzzy number
A generalized fuzzy number
A (a, b, c, d; w) is described as any fuzzy subset of the real line , whose
membership function   A
μ x satisfies the following conditions:
i.   A
μ x is a continuous mapping from  to the closed interval  0, ω , 0 ω 1  ,
ii.     A
μ x = 0, for all x - , a  ,
iii.     L A
μ x L x is strictly increasing on  a, b ,
iv.     A
μ x ω, for all b, c and ω is a constant, 0 < ω 1  ,
v.     R A
μ x R x is strictly decreasing on  c, d ,
vi.     A
μ x 0, for all x d,   where a, b, c, d are real numbers such that a < b c < d .
Definition 2.2. A fuzzy set A is called normal fuzzy set if there exists an element (member) ‘x’ such that
  A
μ x 1 . A fuzzy set A is called convex fuzzy set if
          1 2 1 2A A A
μ αx + 1 - α x min μ x , μ x where  1 2x , x X and α 0, 1  . The set

   α
A
A x X μ x α   is said to be the α - cut of a fuzzy set A .
Definition 2.3. A fuzzy subset A of the real line  with membership function   A
μ x such that
    A
μ x : 0, 1 , is called a fuzzy number if A is normal, A is fuzzy convex,   A
μ x is upper semi-
continuous and  Supp A is bounded, where      A
Supp A cl x : μ x 0   and ‘cl’ is the closure
operator.
It is known that for a normalized tfn. A (a, b, c, d; 1) , there exists four numbers a, b, c, d   and two
functions        A A
L x , R x : 0, 1 , where      A A
L x and R x are non-decreasing and non-
increasing functions respectively. And its membership function is defined as follows:
     A A
μ x {L x =(x-a)/(b-a) for a x b; 1 for b x c;       A
R x =(x-d)/(c-d) for c x d 
and 0 otherwise}. The functions   A
L x and   A
R x are also called the left and right side of the fuzzy
number A respectively [17]. In this paper, we assume that   A x dx < +


 and it is known that the
α - cut of a fuzzy number is 
     α
A
A x μ x α , for α 0, 1    and  
 
0 α
α 0, 1
A = cl A

 
  
 
 ,
according to the definition of a fuzzy number, it is seen at once that every α - cut of a fuzzy number is a
closed interval. Hence, for a fuzzy number A , we have         L UA α A α , A α    where
     L A
A α inf x : μ x α   and      U A
A α sup x : μ x α   . The left and right sides
of the fuzzy number A are strictly monotone, obviously, LA and UA are inverse functions of   A
L x and
  A
R x respectively. Another important type of fuzzy number was introduced in [11] as follows:
Let a, b, c, d   such that a < b c < d . A fuzzy number A defined as     A
μ x : 0, 1 ,

  
n n
A
x - a d - x
μ x for a x b; 1 for b x c; for c x d; 0 otherwise
b - a d - c
   
         
   
where
n > 0, is denoted by   n
A a, b, c, d . And  
n
x - a
L x
b - a
 
  
 
and  
n
d - x
R x
d - c
 
  
 
can also be
termed as left and right spread of the tfn. [Dubois and Prade in 1981].
If   n
A a, b, c, d , then[1-4],
            n n
α L UA A α , A α a + b - a α, d - d - c α ; α 0, 1        .
When n = 1 and b = c, we get a triangular fuzzy number. The conditions r = 1, a = b and c = d imply
the closed interval and in the case r = 1, a = b = c = d = t (some constant), we can get a crisp number ‘t’.
Since a trapezoidal fuzzy number is completely characterized by n = 1 and four real numbers a b c d   ,
it is often denoted as   A a, b, c, d . And the family of trapezoidal fuzzy numbers will be denoted
by  T
F  . Now, for n = 1we have a normal trapezoidal fuzzy number   A a, b, c, d and the
corresponding α - cut is defined by
      αA a + α b - a , d - α d - c ; α 0, 1 (2.4)       . And we need the following results which
can be found in [22, 24].
Result 2.1. Let   D = a, b , a b and a, b  , the set of all closed, bounded intervals on the real line
.
Result 2.2. Let    A = a, b and B = c, d be in D . Then A = B if a = c and b = d .
III. TWO – SAMPLE (STUDENT) t-TEST
Let i jx and y , i = 1, 2, ..., m and j = 1, 2, ..., n be two random samples from two different
normal populations with sizes m and n such that m + n 30 . And let x and y be the mean values and
1 2s and s be the sample standard deviations of the random variables i jx and y respectively and they are
given by [21],
m n
i i
i = 1 i = 1
1 1
x = x and y = y
m n
   
   
   
 
   
m n2 2
1 i 2 j
i = 1 j = 1
1 1
s = x x and s = y y (1)
m - 1 n - 1
     
         
      
 
Let 1μ and 2μ be the population means of X-sample and Y-sample respectively. In testing the null
hypothesis 0 1 2H :μ = μ assuming with equal population standard deviation, we generally use the test statistic:
x - y
t =
1 1
s
m n

where
   2 2
1 2m - 1 s n - 1 s
s =
m + n - 2

In testing the null hypothesis 0 1 2H :μ = μ assuming with unequal population standard deviation, we
commonly use the test statistic:
2 2
1 2
x - y
t =
s s
m n

Now, the degrees of freedom used in this test is n + m - 2v  . Let α be the level of significance and

let α,t v be the tabulated value of ‘t’ for v degrees of freedom at α level of significance. And the null hypothesis
is given by 0 1 2H :μ = μ . And the rejection region for the alternative hypothesis AH at α level is given
below:
If α, m + n - 2t < t (one tailed test), the difference between 1 2μ and μ is not significant at α level.
Then the means of the populations are identical. That is, 1 2μ = μ at α level of significance. Therefore, the null
hypothesis 0H is accepted. Otherwise, the alternative hypothesis AH is accepted.
If α , m + n - 2
2
t < t (two tailed test), the difference between 1 2μ and μ is not significant at α level.
Then the means of the populations are identical. That is, 1 2μ = μ at α level of significance. Therefore, the null
hypothesis 0H is accepted. Otherwise, the alternative hypothesis AH is accepted.
Now, the  100 1 - α % confidence limits for the difference of population means
1 2μ and μ corresponding to the given samples are given by,
     α 1 2 α, m + n - 2 , m + n - 2
2 2
1 1 1 1
x - y t s μ - μ x + y + t s
m n m n
   
          
   
(for equal standard deviations) Or
     
2 2 2 2
1 2 1 2
α 1 2 α, m + n - 2 , m + n - 2
2 2
s s s s
x - y t μ - μ x + y + t
m n m n
   
       
   
   
(for unequal standard deviations)
IV. TEST OF HYPOTHESIS FOR INTERVAL DATA
Let   i ia , b , i = 1, 2, ..., m be a random small sample (X-sample) with size m and
 j jc , d , j = 1, 2, ..., n   be a random small sample (Y-sample) with size n. And let  1 1η , μ be mean of
X from a normal population and let  2 2η , μ be mean of Y from another normal population.
Now, we test the null hypothesis 0H such that the means of the population of the given samples are
equal. That is    0 1 1 2 2 1 2 1 2H : η , μ = η , μ η = η and μ = μ . And the alternative hypotheses are given
by [16],
i.    A 1 1 2 2 1 2 1 2H : η , μ < η , μ η < η and μ < μ
ii.    A 1 1 2 2 1 2 1 2H : η , μ > η , μ η > η and μ > μ
iii.    A 1 1 2 2 1 2 1 2H : η , μ η , μ η η or μ μ   
Now the lower values and upper values for X-sample and Y-sample are given below:
Alternative Hypothesis AH Rejection Region at α Level
A 1 2H :μ > μ α, m + n - 2t t (Upper tailed test)
A 1 2H :μ < μ α, m + n - 2t - t (Lower tailed test)
A 1 2H :μ μ α , m + n - 2
2
t t (Two tailed test)
LX (Lower values of X-sample) ia ;i = 1, 2, ..., m
LY (Lower values of Y-sample) jc ; j = 1, 2, ..., n
UX (Upper values of X-sample) ib ;i = 1, 2, ..., m
UY (Upper values of Y-sample) jd ; j = 1, 2, ..., n

Let Lx and Ly be the sample means, Lxs and Lys be the sample standard deviation of
L
X and
L
Y respectively. Similarly let Ux and Uy be the sample means, Uxs and Uys be the sample standard
deviation of
U
X and
U
Y respectively.
Case (i): If the population standard deviations are assumed to be equal, then under the null hypothesis
   0 1 1 2 2H : η , μ = η , μ , the test statistic is given by,
L L
L
L
x - y
t =
1 1
s
m n

and
U U
U
U
x - y
t =
1 1
s
m n

where
   L L
2 2
x y
L
m - 1 s n - 1 s
s =
m + n - 2

and
   U U
2 2
x y
U
m - 1 s n - 1 s
s =
m + n - 2

Case (ii): If the population standard deviations are assumed to be unequal, then under the null
hypothesis    0 1 1 2 2H : η , μ = η , μ , the test statistic is given by,
L L
L L
L 2 2
x y
x - y
t =
s s
m n

and
U U
U U
U 2 2
x y
x - y
t =
s s
m n

where the standard deviations for upper and lower values of the samples of X and Y are given by the equation
(1).
And the rejection region of the alternative hypothesis AH at α level of significance is given below:
If L α, m + n - 2t < t and U α, m + n - 2t < t (one tailed test), then the difference between  1 1η , μ and
 2 2η , μ is not significant at α level. Then the means of the populations are identical. That is,
   1 1 2 2η , μ = η , μ at α level of significance. Therefore, the null hypothesis 0H is accepted. Otherwise, the
alternative hypothesis AH is accepted.
If L α , m + n - 2
2
t < t and U α , m + n - 2
2
t < t (two tailed test), the difference between  1 1η , μ and
 2 2η , μ is not significant at α level. Then the means of the populations are identical. That
is,   1 1 2 2η , μ = η , μ at α level of significance. Therefore, the null hypothesis 0H is accepted. Otherwise,
the alternative hypothesis AH is accepted.
And the  100 1 - α % confidence limits for the difference of lower limit and upper limit of the population
means  1 1η , μ and  2 2η , μ corresponding to the given samples are given below:
     L Lα L 1 2 α LL L, m + n - 2 , m + n - 2
2 2
1 1 1 1
x - y t s η - η x - y + t s
m n m n
   
          
   
(for equal population standard deviations), and
Alternative Hypothesis AH Rejection Region at α Level
   A 1 1 2 2H : η , μ > η , μ L α, m + n - 2t t and U α, m + n - 2t t (Upper tailed test)
   A 1 1 2 2H : η , μ < η , μ L α, m + n - 2t -t and U α, m + n - 2t -t (Lower tailed test)
   A 1 1 2 2H : η , μ η , μ L α , m + n - 2
2
t t or U α , m + n - 2
2
t t (Two tailed test)

     U Uα U 1 2 α UU U, m + n - 2 , m + n - 2
2 2
1 1 1 1
x - y t s η - η x - y + t s
m n m n
   
          
   
(for equal population standard deviations) Or
     L L L L
2 2 2 2
x y x y
L Lα 1 2 αL L, m + n - 2 , m + n - 2
2 2
s s s s
x - y t η - η x - y + t
m n m n
   
       
   
   
(for unequal population standard deviations) and
     U U U U
2 2 2 2
x y x y
U Uα 1 2 αU U, m + n - 2 , m + n - 2
2 2
s s s s
x - y t η - η x - y + t
m n m n
   
       
   
   
(for unequal population standard deviations)
Decision table:
Partial acceptance of null hypothesis H0 at the intersection of certain level of  at both upper level and lower
level models can be taken into account for the acceptance of the null hypothesis 0H .
This test procedure has been illustrated using the following numerical examples.
Example-1
The following interval data are given the gain in weights (in lbs) of pet dogs fed on two kinds of diets
A and B [21].
Now, we test if the two diets differ significantly on the basis of their nutrition effects on increase in the weight
of the pet dogs.
Here the null hypothesis is,    0 1 1 2 2H : η , μ = η , μ 1 2 1 2η = η and μ = μ .
There is no significant difference between the nutrition effects from diet A and diet B.
Acceptance of null hypotheses 0H
Lower Level Model Upper Level Model Conclusion
If H0 is accepted for all
 α [0,1]
and H0 is accepted for all
 α [0,1]
then 0H is accepted for all
 α [0,1]
If H0 is accepted for all
 α [0,1]
and H0 is rejected for all
 α [0,1]
then 0H is rejected for all
 α [0,1]
If H0 is rejected for all
 α [0,1]
and H0 is accepted for all
 α [0,1]
then 0H is rejected for all
 α [0,1]
If H0 is rejected for all
 α [0,1]
or H0 is rejected for all  α [0,1] then 0H is rejected for all
 α [0,1]
Diet-A Diet-B Diet-A Diet-B
 18, 19  22, 26  19, 22  22, 28
 16, 18  27, 31  20, 24  20, 24
 30, 32  25, 28  27, 30  11, 15
 28, 30  12, 16  18, 22  14, 17
 22, 24  16, 20  21, 24  17, 21
 14, 16  18, 22 --  25, 27
 28, 32  26, 30 --  19, 22
--  23, 25

And the alternative hypothesis is    A 1 1 2 2H : η , μ η , μ 1 2 1 2η η or μ μ   (Two tailed test).
 The two kinds of the diets differ significantly on the basis of their nutrition effects.
We assume that the standard deviations of the populations are not equal and we use 5% level of significance.
Here m=12 and n=15.
The tabulated value of ‘t’ for m + n – 2 = 27 – 2 = 25 degrees of freedom at 5% level of significance is
αT 2.06
Now, L
m
L i
i = 1
1
x = x
m
 
 
 
  Lx 21.75 and L
n
iL
i = 1
1
y = y
n
 
 
 
  Ly 20.0667 and
U
m
U i
i = 1
1
x = x
m
 
 
 
  Ux 24.4167 and U
n
iU
i = 1
1
y = y
n
 
 
 
  Uy 23.7333
 L L
m 2
2
Lx i
i = 1
1
s x -x
m - 1
 
  
 
 L
2
xs 27.8409  and  L L
n 2
2
y i L
i = 1
1
s y -y
n - 1
 
  
 
 L
2
ys 27.0667 
 U U
m 2
2
Ux i
i = 1
1
s x -x
m - 1
 
  
 
 U
2
xs 30.0833  and  U U
n 2
2
y i U
i = 1
1
s y -y
n - 1
 
  
 
 U
2
ys 26.6381 
The Test Statistics:
L L
L L
L 2 2
x y
x - y
t =
s s
m n

Lt 0.8288  and
U U
U U
U 2 2
x y
x - y
t =
s s
m n

Ut 0.3302 
Since, L αt T 2.06  and U αt T 2.06  , we accept the null hypothesis 0H .
 There is no significant difference between the nutrition effects of the diets A and B at 5% level of
significance.
V. TEST OF HYPOTHESIS FOR FUZZY DATA USING TFNS.
Definition 5.1: Trapezoidal Fuzzy Number to Interval
Let a trapezoidal fuzzy number be defined as   A a, b, c, d , then the fuzzy interval in terms of α -
cut interval is defined as follows [33]:
    αA a + b - a α, d - d - c α ; 0 α 1 (5.1)       
Suppose that the given sample is a fuzzy data that are trapezoidal fuzzy numbers and we have to test
the hypothesis about the population mean. Using the relation (1) and the proposed test procedure, we can test the
hypothesis by transferring the fuzzy data into interval data.
Example-2
Two kinds of engine oils A and B for automobiles are under mileage test for some taxies, then we
request the taxi drivers to record the consumption of fuel. Due to limited available source, the data are recorded
as trapezoidal fuzzy numbers which are given in the following table. Suppose the random variables have normal
distribution and their variances of both populations are known and equal with one. We now investigate the
effects of the two kinds of engine oils on consumption of fuel at 5% level of significance [10].
A B
 4, 4.5, 5, 6  5, 6.5, 7, 8
 3.5, 4, 5, 6.5  4, 4.5, 5, 6
 5, 5.5, 5.8, 6  5.5, 7, 8, 8.5
 5.5, 5.8, 6, 6.5  5, 6, 6.5, 7
 3, 3.5, 4, 5  6, 6.5, 7, 8
--  6, 7.5, 8.5, 9

Now the interval representation of the above trapezoidal data is given below:
αA 
 
αB 
 
 4 + 0.5α, 6 - α  5 + 1.5α, 8 - α
 3.5 + 0.5α, 6.5 - 1.5α  4 + 0.5α, 6 - α
 5 + 0.5α, 6 - 0.2α  5.5 + 1.5α, 8.5 - 0.5α
 5.5 + 0.3α, 6.5 - 0.5α  5 + α, 7 - 0.5α
 3 + 0.5α, 5 - α  6 + 0.5α, 8 - α
--  6 + 1.5α, 9 - 0.5α
Lower Level Samples Upper Level Samples
Lx Ly Ux Uy
4 + 0.5α 5 + 1.5α 6 - α 8 - α
3.5 + 0.5α 4 + 0.5α 6.5 - 1.5α 6 - α
5 + 0.5α 5.5 + 1.5α 6 - 0.2α 8.5 - 0.5α
5.5 + 0.3α 5 + α 6.5 - 0.5α 7 - 0.5α
3 + 0.5α 6 + 0.5α 5 - α 8 - α
-- 6 + 1.5α -- 9 - 0.5α
Here, m = 5 and n = 6.
L
m
L i
i = 1
1
x = x
m
 
 
 
 Lx 4.2 0.46α   and U
m
U i
i = 1
1
x = x
m
 
 
 
 Ux 6 0.84α  
L
n
iL
i = 1
1
y = y
n
 
 
 
 Ly 5.25 1.083α   and U
n
iU
i = 1
1
y = y
n
 
 
 
 Uy 7.75 0.75α  
 L L
m 2
2
Lx i
i = 1
1
s x -x
m - 1
 
  
 
 2
0.008α 0.13α + 1.075  and
 L L
n 2
2
y i L
i = 1
1
s y -y
n - 1
 
  
 
 2
0.2417α 0.25α + 0.5750 
 U U
m 2
2
Ux i
i = 1
1
s x -x
m - 1
 
  
 
 2
0.2402α 0.375  and
 U U
n 2
2
y i U
i = 1
1
s y -y
n - 1
 
  
 
 2
0.05α 0.25α - 0.05  and
   
 
L L
2 2
x y2
L
m - 1 s n - 1 s
S
m + n - 2

 2
0.1379α 0.1967α + 0.7972 
   
 
U U
2 2
x y2
U
m - 1 s n - 1 s
S
m + n - 2

 2
0.1346α 0.1389α + 0.1389 
Null hypothesis:
  0H :    The two kinds of engine oils on fuel consumption are same.
Alternative hypothesis:
  AH :    The two kinds of engine oils on fuel consumption differ significantly.

Here,   1 1η , μ    and   2 2η , μ    .
And therefore,   0H :            
0 1 2H : η = η and 1 2μ = μ .
  AH :            
A 1 2H : η η  or 1 2μ μ (Two tailed test).
Now, the tabulated value of ‘t’ at 5% level of significance with 9 degrees of freedom is αT 2.262
Test statistics:
L L
L L
L 2 2
x y
1.9422 if α = 0
2.0814 if α = 0.1
2.2204 if α = 0.2x - y
t =
2.3578 if α = 0.3s s
...m n
3.2154 if α = 1



 

 


L αt T  for 0.3 α 1 
U U
U U
U 2 2
x y
-7.7537 if α = 0x - y
t =
-4.7320 if α = 1s s
m n

 


U αt T  for 0 α 1 
Conclusion
Hence, L αt T and U αt T for 0.3 α 1  implies the null hypothesis 0H is rejected and we
accept the alternative hypothesis AH . Therefore, the two kinds of engine oils for automobiles on consumption
of fuel are not the same at 5% level of significance.
Remark
The obtained result from the above test procedure in Example-2 differs by the lower value of α by 0.3
when compared with the result in Baloui Jamkhaneh and Nadi Gara [10] which is 0 α 1  when performing
this test procedure using trapezoidal fuzzy interval data.
VI. WANG’S CENTROID POINT AND RANKING METHOD
Wang et al. [37] found that the centroid formulae proposed by Cheng are incorrect and have led to some
misapplications such as by Chu and Tsao. They presented the correct method for centroid formulae for a
generalized fuzzy number   A= a, b, c, d; w as
   
       
0 0
1 dc - ab w c - b
x , y a + b + c + d , 1 (6.1)
3 d + c - a + b 3 d + c - a + b
        
                            
And the ranking function associated with A is  
2 2
0 0R A x + y (6.2) 
For a normalized tfn, we put w = 1 in equations (6.1) so we have,
   
       
0 0
1 dc - ab 1 c - b
x , y a + b + c + d , 1 (6.3)
3 d + c - a + b 3 d + c - a + b
        
                            
And the ranking function associated with A is  
2 2
0 0R A x + y (6.4)  .
Let  i jA and A be two fuzzy numbers,      i j i j(i) R A R A then A A     i j(ii) R A R A

 i jthen A A and      i j i j(iii) R A =R A then A A .
Two-sample t-test using Wang’s centroid point and ranking function
Example 6.1. Let we consider example 2, using the above relations (6.3) and (6.4), we obtain the ranks of tfns.
which are tabulated below:
Here, the mean values are   A
R x 5.0394 ,   B
R y 6.5812 and the variance values are
2
xs 0.6204 ,
2
ys 0.9611 and the calculated value of ‘t’ is t = -2.8919. The tabulated value of ‘t’ at 5% level of
significance with 9 degrees of freedom is αT 2.262 . And αt T (two tailed test).The null hypothesis
0H is rejected. Therefore, the two kinds of engine oils for automobiles on consumption of fuel are not
the same at 5% level of significance.
VII. REZVANI’S RANKING FUNCTION OF TFNS.
The centroid of a trapezoid is considered as the balancing point of the trapezoid. Divide the trapezoid into three
plane figures. These three plane figures are a triangle (APB), a rectangle (BPQC) and a triangle (CQD)
respectively. Let the centroids of the three plane figures be 1 2 3G , G and G respectively. The incenter of
these centroids 1 2 3G , G and G is taken as the point of reference to define the ranking of generalized
trapezoidal fuzzy numbers. The reason for selecting this point as a point of reference is that each centroid point
are balancing points of each individual plane figure and the incenter of these centroid points is much more
balancing point for a generalized trapezoidal fuzzy number. Therefore, this point would be a better reference
point than the centroid point of the trapezoid.
Consider a generalized trapezoidal fuzzy number   A= a, b, c, d; w . The centroids of the three plane figures
are:
1 2 3
a+2b w b+c w 2c+d w
G , , G , and G , (7.1)
3 3 2 2 3 3
     
          
     
Equation of the line 1 3G G is
w
y =
3
and 2G does not lie on the line 1 3G G . Therefore, 1 2 3G , G and G
are non-collinear and they form a triangle. We define the incenter  0 0I x , y of the triangle with vertices
1 2 3G , G and G of the generalized fuzzy number   A= a, b, c, d; w as [30]
  iA
R x ; i=1, 2, ..., 5   jB
R y ; j=1, 2, ..., 6
4.9163 6.6062
4.8097 4.9163
5.5766 7.2204
5.9738 6.1131
3.9205 6.9116
-- 7.7196

  0 0A
a+2b b+c 2c+d w w w
α β γ α β γ
3 2 3 3 2 3
I x , y , (7.2)
α + β + γ α + β + γ
            
               
               
 
  
     
2 2 22 2
c - 3b + 2d w 2c + d - a - 2b 3c - 2a - b w
where α ,β ,γ
6 3 6
 
  
And ranking function of the trapezoidal fuzzy number   A= a, b, c, d; w which maps the set of all fuzzy
numbers to a set of all real numbers i.e. R: A     
 is defined as  
2 2
0 0R A x + y (7.3)   
which is the Euclidean distance from the incenter of the centroids. For a normalized tfn. we put w = 1 in
equations (7.1), (7.2) and (7.3) so we have,
1 2 3
a+2b 1 b+c 1 2c+d 1
G , , G , and G , (7.4)
3 3 2 2 3 3
     
          
     
  0 0A
a+2b b+c 2c+d 1 1 1
α β γ α β γ
3 2 3 3 2 3
I x , y , (7.5)
α + β + γ α + β + γ
            
               
               
 
  
     
2 2 2
c - 3b + 2d 1 2c + d - a - 2b 3c - 2a - b 1
where α ,β and γ
6 3 6
 
  
And ranking function of the trapezoidal fuzzy number   A= a, b, c, d; 1 is defined
as  
2 2
0 0R A x + y (7.6)    .
Two-sample t-test using Rezvani’s ranking function
We now analyse the one-sample t-test by assigning rank for each normalized trapezoidal fuzzy numbers and
based on the ranking grades the decisions are observed.
Example 7.1. Let we consider example 2, using the above relations (7.4), (7.5) and (7.6), we obtain the ranks
of tfns. which are tabulated below:
R x 4.9299 ,   B
2
xs 0.7573 ,
2
0H is rejected. Therefore, the two kinds of engine oils for automobiles on consumption of fuel are not
the same at 5% level of significance.
  iA
R x ; i=1, 2, ..., 5   jB
R y ; j=1, 2, ..., 6
4.7724 6.7603
4.5225 4.7724
5.6571 7.5082
5.9203 6.2594
3.7772 6.7671
-- 8.0074

VIII. GRADED MEAN INTEGRATION REPRESENTATION (GMIR)
Let   A= a, b, c, d; w be a generalized trapezoidal fuzzy number, then the GMIR [29] of A is defined
by      -1 -1w w
0 0
L h R h
P A h dh / hdh
2
 
  
 
  .
Theorem 8.1.Let   A= a, b, c, d; 1 be a tfn. with normal shape function, where a, b, c, d are real numbers
such that a < b c < d . Then the graded mean integration representation (GMIR) of
A is    
 
a + d n
P A b - a - d + c
2 2n + 1
  .
Proof :For a trapezoidal fuzzy number   n
A= a, b, c, d; 1 , we have  
n
x - a
L x
b - a
 
  
 
and
 
n
d - x
R x
d - c
 
  
 
.
Then,    
n
1
-1 n
x - a
h = L h a + b - a h
b - a
 
  
 
;
   
n
1
-1 n
d - x
h = R h d - d - c h
d - c
 
  
 
       
1 1
1 1
n n
0 0
1
P A h a + b - a h d - d - c h d h / h d h
2
       
  
 
   a + d1 n 1= b - a - d + c /
22 2 2 n + 1
  
   
  
   
 
a + d n
Thus, P A b - a - d + c
2 2n + 1
  Hence the proof.
Result 8.1.If n =1 in the above theorem, we have   a + 2b + 2c + d
P A
6

Two-sample t-test using GMIR of tfns.
Example 8.1. Let us consider example 2, using the result-8.1 from above theorem-8.1, we get the GMIR of
each tfns. ix , i=1, 2, ..., 5; 
jy , j=1, 2, ..., 6 which are tabulated below:
P x 4.9733 ,   B
P y 6.6111 and the variance values are
2
xs 0.6819, 2
ys 1.0852 and the calculated value of ‘t’ is t = -2.9078. The tabulated value of ‘t’ at 5%
level of significance with 9 degrees of freedom is αT 2.262 . And αt T (two tailed test). The null
hypothesis 0H is rejected. Therefore, the two kinds of engine oils for automobiles on consumption of fuel
are not the same at 5% level of significance.
  iA
P x ; i=1, 2, ..., 5   jB
P y ; j=1, 2, ..., 6
4.8333 6.6667
4.6667 4.8333
5.6 7.3333
5.9333 6.1667
3.8333 6.8333
-- 7.8333

IX. TWO-SAMPLE t-TEST USING TOTAL INTEGRAL VALUE (TIV) OF TFNS.
The membership grades for a normalized tfn.   A a, b, c, d; 1 is calculated by the relation
[23] 
 
 
b c d
A
a b cSupp A
x - a x - d
μ x dx = dx + dx + dx (9.1)
b - a c - d
   
     
   
   
Example 9.1. Let us consider example 2, the total integral value of first entry   1A 4, 4.5, 5, 6 will be

 
 1
1
4.5 5 6
A
4 4.5 5Supp A
x - 4 x - 6
μ x dx = dx + dx + dx =1.25 I
0.5 1
   
   
   
   
The total integral values of remaining entries can be calculated in similar way, which have been tabulated
below:
I x 1.15 ,   B
I y 1.5833 and the variance values are
2
xs 0.32375, 2
ys 0.141667 and the calculated value of ‘t’ is t = -1.4578. The tabulated value of ‘t’ at 5%
level of significance with 9 degrees of freedom is αT 2.262 . And αt < T (two tailed test). The null
hypothesis 0H is accepted. Therefore, the two kinds of engine oils for automobiles on consumption of
fuel do not differ significantly at 5% level of significance.
X. LIOU AND WANG’S CENTROID POINT METHOD
Liou and Wang [25] ranked fuzzy numbers with total integral value. For a fuzzy number defined by definition
(2.3), the total integral value is defined as
       α
i i iT R LI x αI x 1 - α I x (10.1)     where
    
 
i
i
d
iR x
cSupp x
x-d
I x R x dx= dx---(10.2)
c-d
 
  
 
  and
    
 
i
i
b
iL x
aSupp x
x-a
I x L x dx= dx ---(10.3)
b-a
 
  
 
  are the right and left integral values respectively and
0 α 1  .
(i)  α 0,1 is the index of optimism which represents the degree of optimism of a decision maker. (ii) If
α 0 , then the total value of integral represents a pessimistic decision maker’s view point which is equal to
left integral value. (iii) If α 1 , then the total integral value represents an optimistic decision maker’s view
point and is equal to the right integral value.(iv) If α 0.5 then the total integral value represents a moderate
decision maker’s view point and is equal to the mean of right and left integral values. For a decision maker,
the larger the value of α is, the higher is the degree of optimism.
Two-sample t-test using Liou and Wang’s centroid point method:
Example 10.1. Let we consider example 2, using the above equations (10.1), (10.2) and (10.3), we get the
centroid point of first member as follows:

 
 iA
Supp A
μ x dx = I; i = 1, 2, ..., 5 
 
 jB
Supp B
μ y dy = I; j = 1, 2, ..., 6
1.25 1.75
2 1.25
0.65 2
0.6 1.25
1.25 1.25
-- 2

   
4.5 6
1 1L R
4 5
x - 4 x - 6
I x dx 1/ 4 ; I x dx 1/ 2
0.5 1
   
      
   
  .  α
1TTherefore I x (1 α) / 4  .
Similarly we can find  α
iTI x ; for i = 2, ..., 5;  α
T jI y ; for j = 1, 2, ..., 6 and the calculated values are
tabulated below:
α
TI x (23 19α) /100  ,   B
α
TI y (13 4α) / 24  and the variance values
are
2
xs 0 ,
2
ys 0 . As the variance values vanish for both the sample data x and y, we can not decide
any conclusion for this case using Liou and Wang’s method.
XI. THORANI’S RANKING METHOD
As per the description in Salim Rezvani’s ranking method, Thorani et al. [31] presented a different kind of
centroid point and ranking function of tfns. The incenter   0 0A
I x , y of the triangle [Fig. 1] with vertices
1 2 3G , G and G of the generalized tfn.   A= a, b, c, d; w is given by,
  0 0A
a+2b b+c 2c+d w w w
α β γ α β γ
3 2 3 3 2 3
I x , y , (11.1)
α + β + γ α + β + γ
            
               
               
 
  
     
2 2 22 2
c - 3b + 2d w 2c + d - a - 2b 3c - 2a - b w
where α ,β ,γ
6 3 6
 
  
And the ranking function of the generalized tfn.   A= a, b, c, d; w which maps the set of all fuzzy numbers
to a set of real numbers is defined as   0 0R A x y (11.2)     . For a normalized tfn., we put w = 1 in
equations (11.1) and (11.2) so we have,
  0 0A
a+2b b+c 2c+d 1 1 1
α β γ α β γ
3 2 3 3 2 3
I x , y , (11.3)
α + β + γ α + β + γ
            
               
               
 
  
     
2 2 2
c - 3b + 2d 1 2c + d - a - 2b 3c - 2a - b 1
where α ,β and γ
6 3 6
 
  
And for   A= a, b, c, d; 1 ,   0 0R A x y (11.4)    
Two-sample t-test using Thorani’s ranking method of tfns.
Example 11.1. Let us consider example 2, using the above relations (11.3) and (11.4), we get the ranks of each
tfns.  i jx , i =1, 2, ..., 5; y , j =1, 2, ..., 6 which are tabulated below:
 α
iTI x ; for i = 1, 2, ..., 5  α
T jI y ; for j = 1, 2, ..., 6
(1+)/4 (3-)/4
(1+2)/4 (1+)/4
(5-3)/20 (3-2)/4
(3+2)/20 (2-)/4
(1+)/4 (1+)/4
-- (3-2)/4

R x 2.0240 ,   B
2
xs 0.1191 ,
2
0H is rejected. Therefore, the two kinds of engine oils for automobiles on consumption of fuel differ
significantly at 5% level of significance.
XII. GENERAL CONCLUSION
The decisions obtained from various methods are tabulated below for the acceptance of null hypothesis.
Observing the decisions obtained from α-cut interval method, for example-2, the null hypothesis is rejected for
the level of 0.3 α 1  at both l.l.m. and u.l.m. On the other hand, using Liou & Wang’s method (L&W), we
cannot conclude any decision as the variance values are zero. Moreover, the method of total integral value of
tfns. (TIV) do not provide reliable results as it accepts the null hypothesis 0H while other methods are rejecting
0H and the decisions obtained from Wang’s ranking method, Rezvani’s ranking method, GMIR and Thorani’s
ranking method of tfns. provide parallel discussions.
REFERENCES
[1] S. Abbasbandy, B. Asady, The nearest trapezoidal fuzzy number to a fuzzy quantity, Appl. Math. Comput., 156 (2004) 381-386.
[2] S. Abbasbandy, M. Amirfakhrian, The nearest approximation of a fuzzy quantity in parametric form, Appl. Math. Comput., 172
(2006) 624-632.
[3] S. Abbasbandy, M. Amirfakhrian, The nearest trapezoidal form of generalized left right fuzzy number, Internat. J. Approx.
Reason., 43 (2006) 166-178.
[4] S. Abbasbandy, T. Hajjari, Weighted trapezoidal approximation-preserving cores of a fuzzy number, Computers and
Mathematics with Applications, 59 (2010) 3066-3077.
[5] Abhinav Bansal, Trapezoidal Fuzzy Numbers (a, b, c, d): Arithmetic Behavior, International Journal of Physical and
Mathematical Sciences (2011), 39-44.
[6] M. G. Akbari and A. Resaei, Bootstrap statistical inference for the variance based on fuzzy data. Austrian Journal of Statistics,
38 (2009), 121-130.
[7] M. Arefi and S. M. Taheri, Testing fuzzy hypotheses using fuzzy data based on fuzzy test statistic. Journal of Uncertain
Systems, 5 (2011), 45-61.
[8] B. F. Arnold, Testing fuzzy hypotheses with crisp data, Fuzzy Sets and Systems 94 (1998), 323-333.
[9] Asady, B. Trapezoidal Approximation of a Fuzzy Number Preserving the Expected Interval and Including the Core, American
Journal of Operations Research, 3 (2013), 299-306.
[10] E. Baloui Jamkhaneh and A. Nadi Ghara, Testing statistical hypotheses for compare means with vague data, International
Mathematical Forum, 5 (2010) 615-620.
[11] S. Bodjanova, Median value and median interval of a fuzzy number, Inform. Sci. 172 (2005) 73-89.
[12] James J. Buckley, Fuzzy Probability and Statistics, Springer-Verlag Berlin Heidelberg 2006.
[13] M. R. Casals, M. A. Gil and P. Gil, The fuzzy decision problem: an approach to the problem of testing statistical hypotheses with
fuzzy information, European Journal of Operational Research 27 (1986), 371-382.
[14] M. R. Casals, M. A. Gil, A note on the operativeness of Neyman-Pearson tests with fuzzy information, Fuzzy Sets and Systems
30 (1989) 215-220.
[15] J. Chachi, S. M. Taheri and R. Viertl, Testing statistical hypotheses based on fuzzy confidence intervals, Forschungsbericht SM-
2012-2, Technische Universitat Wien, Austria, 2012.
[16] J. L. Devore, Probability and Statistics for Engineers, Cengage, 2008.
[17] D. Dubois and H. Prade, Operations on fuzzy numbers, Int. J. Syst. Sci., 9 (1978) 613-626.
  iA
R x ; i=1, 2, ..., 5   jB
R y ; j=1, 2, ..., 6
1.9703 2.8028
1.8726 1.9703
2.3114 3.1172
2.4098 2.5883
1.5559 2.7991
-- 3.3251
Acceptance of null hypothesis 0H at 5% level of significance
α-cut Wang Rezvani GMIR TIV L&W Thorani
    
Cannot be
decided


[18] P. Gajivaradhan and S. Parthiban, Statistical hypothesis testing through trapezoidal fuzzy interval data, Int. Research J. of Engg.
and Tech. Vol. 02, Issue: 02, May 2015, pp. 251-258.
[19] P. Gajivaradhan and S. Parthiban, Two sample statistical hypothesis test for trapezoidal fuzzy interval data, Int. J. of Applied
Mathematics and Statistical Sciences. Vol. 04, Issue: 05, Aug.-Sep. 2015, pp. 11-24.
[20] P. Grzegorzewski, Testing statistical hypotheses with vague data, Fuzzy sets and Systems, 112 (2000), 501-510.
[21] S. C. Gupta and V. K. Kapoor, Fundamentals of Mathematical Statistics (A Modern Approach), Sultan Chand & Sons, New
Delhi.
[22] R. R. Hocking, Methods and applications of linear models: regression and the analysis of variance, New York: John Wiley &
Sons, 1996.
[23] Iuliana Carmen BĂRBĂCIORU, Statistical Hypothesis Testing Using Fuzzy Linguistic Variables, Fiabilitatesi Durabilitate-
Fiability & Durability, Supplement, 1 (2012) Editura “Academica Brȃncusi”, Tȃrgu Jiu, ISSN 1844-640X.
[24] George J. Klir and Bo Yuan, Fuzzy sets and fuzzy logic, Theory and Applications, Prentice-Hall, New Jersey, 2008.
[25] T. S. Liou and M. J. Wang, Ranking Fuzzy Numbers with Integral Value, Fuzzy Sets and Systems, 50 (1992) 247-225.
[26] J. J. Saade and H. Schwarzlander, Fuzzy hypothesis testing with hybrid data, Fuzzy Sets and Systems, 35 (1990), 197-212.
[27] J. J. Saade, Extension of fuzzy hypothesis testing with hybrid data, Fuzzy Sets and Systems, 63 (1994), 57-71.
[28] S. Salahsour, S. Abbasbandy and T. Allahviranloo, Ranking Fuzzy Numbers using Fuzzy Maximizing-Minimizing points,
EUSFLAT-LFA: July 2011, Aix-les-Bains, France.
[29] Salim Rezvani and Mohammad Molani, Representation of trapezoidal fuzzy numbers with shape function, to appear in Annals of
Fuzzy mathematics and Informatics.
[30] Salim Rezvani, Ranking Generalized Trapezoidal Fuzzy Numbers with Euclidean Distance by the Incentre of Centroids,
Mathematica Aeterna, 3 (2) (2013) 103-114.
[31] Y. L. P. Thorani, et al., Ordering Generalized Trapezoidal Fuzzy Numbers, Int. J. Contemp. Math. Sciences, 7(12) (2012) 555-
573.
[32] J. Ch. Son, I. Song and H. Y. Kim, A fuzzy decision problem based on the generalized Neymen-Pearson criterion, Fuzzy Sets and
Systems, 47 (1992), 65-75.
[33] Superna Das and S. Chakraverty, Numerical Solution of Interval and Fuzzy Systems of Linear Equations, Applications and
Applied Mathematics: An International Journal (AAM): Vol. 7, Issue 1 (June 2012), pp. 334-356.
[34] R. Viertl, Univariate statistical analysis with fuzzy data, Computational Statistics and Data Analysis, 51 (2006), 33-147.
[35] R. Viertl, Statistical methods for fuzzy data, John Wiley and Sons, Chichester, 2011.
[36] N. Watanabe and T. Imaizumi, A fuzzy statistical test of fuzzy hypotheses, Fuzzy Sets and Systems, 53 (1993) 167-178.
[37] Y. M. Wang et al., On the centroids of fuzzy numbers, Fuzzy Sets and Systems, 157 (2006) 919-926.
[38] H. C. Wu, Statistical hypotheses testing for fuzzy data, Information Sciences, 175 (2005), 30-56.
[39] H. C. Wu, Statistical confidence intervals for fuzzy data, Expert Systems with Applications, 36 (2009), 2670-2676.
[40] L. Zadeh, Fuzzy Probabilities, Information Processing and Management 203 (1984), 363-372.

A Comparative Study of Two-Sample t-Test Under Fuzzy Environments Using Trapezoidal Fuzzy Numbers

More Related Content

What's hot (20)

Similar to A Comparative Study of Two-Sample t-Test Under Fuzzy Environments Using Trapezoidal Fuzzy Numbers (20)

Recently uploaded (20)

A Comparative Study of Two-Sample t-Test Under Fuzzy Environments Using Trapezoidal Fuzzy Numbers