Professional Documents
Culture Documents
Research Article
Research Article
Abstract
Inverted exponential distribution (IED) act as life time model that is useful for modeling real life
situations in medical, insurance and engineering. IED is much ambitious as compared to the exponential
distribution as it attains the inverted failure rate models. The properties of IED such as Survival function
and Hazard function are also accompalished.The parameter of IED is estimated by using distinct classical
methods such as Least Square Method (LSM) and Maximum Likelihood Estimator method (MLE).
Elegant Bayes estimators are also derived using informative priors such as Gamma Prior and Exponential
Prior under different symmetric loss functions including Square error loss function (SELF), Quadratic
loss function (QLF) and Entropy loss function (ELF). Finally, the comparison is conducted for simulated
data under the classical and Bayesian estimators. Among these observations it is concluded that Bayesian
approach shows better behavior than classical estimators depending on their Mean square error (MSE)
and Mean absolute percentage error (MAPE). R-software is used to execute the numerical calculations.
Keywords: Inverted Exponential Distribution (IED), Least Square Estimator (LSE), Maximum likelihood
estimator (MLE), Square Error loss function (SELF), Quadratic loss function (QLF), Entropy loss
function (ELF).
1- Introduction
The inverted exponential distribution developed by Keller and Kamath (1982) , is a life testing model. IED is a
member of continuous probability distributions. The major difference between the exponential and IED is that , in
case of ED the interest is to always on the constant failure rate (Lemonte , 2013) that is disadvantage because for
that reason the distribution become unsuitable for modeling real life situations with bathtub and inverted bathtub
failure rate [1]. Rather than IED has the capability to attain the inverted failure rate modeles during the time duration
process. In (2013) Singh et al. discuss the behavior of IED in real life fields e.g. medicine field, business field and
insurance etc [2].
In (1989) Lin et al. appraised the Bayesian form of the classical estimators for IED under complete sample study [3].
In (2015), Al-Noor and Bawi evaluate the estimates for IED under classical and non-classical approach as well [4].
The bayes estimators undergeneral entropy loss function are also derived under Bayesian decision theoretic
approach. The behavior of estimators are also judge under simulation study.
In this paper extensive study conducted on the IED , also explain their relative properties under different aspects.
Parameter estimation conducted under classical and Bayesian approach . We have assumed Inverted gamma prior,
size biased inverted gamma prior , exponential prior and size biased exponential prior taken as their informative
priors. Different symmetric loss functions also take into account to achieve some specific goals.
A continuous r.v ‘X’ is said to be Inverted exponential distribution with parameter ∅ . And their probability density
function and cumulative density functions given as respectively.
f ( x ; ∅ )=¿
The plot for the pdf of IED for various values of parameter ∅ is shown in figure 1;
In this section, presented the reliability function which is also known as survival function. [5] Survival function is an
specialized type of mathematical statistics developed to study a special type of random variable of positive values
with censored observations of which survival time events are most common (Ma and Krings,2008).
S ( x )=Pr ( X > x )
S ( x )=1−Pr ( X ≤ x )
S ( x )=1−F ( x)
Hence reliability function of IE distribution as;
Hazard rate is an important tool of survival analysis that describes the immediate risk of failure at a given time point.
The hazard rate of a random variable X with pdf f (x) and cdf F (x) is [6]:
f (x )
h ( x )=
1−F (x)
Hazard rates will be evaluated for different parameter points (Tiwari et al. 2006).
1
( )
()
2
1 x
h IE ( x )= . (4 )
∅
exp ( )
−1
∅. x
−1
The plot for the hazard function of IED for various values of parameter ∅ is shown in figure 4;
3- Estimation of Parameter
3.1 Maximum Likelihood Estimation
In this section MLE of inverted exponential distribution (∅ ¿ is considered, where ∅ is unknown. If x 1 , x 2 , … … x n
is a random sample from IED (∅ ¿ then likelihood function L ( ∅ ) is given by
L( ∅ , x1 , x2 , … … x n ) = f (x 1 , x 2 , … … x n∨∅ )
n
L ( ∅| x )=∏ f (x i∨ ∅ )
i=1
n n −1
1 1
¿
∅n
∏ 2∏
x i i−1
e ∅ x (5)
i=1
The MLE of ∅ refer as ∅^ MLE obtained by taking derivative of natural log- likelihood function and put it equal to
zero .
n n
1 1
logLf ( x )=−nlog ∅ −2 ∑ log x i− ∑
i=1 ∅ x i=1 i
n
t 1
∅^ MLE=( ) ; Where t=∑
n i=1 xi
∑ {¿ ¿
i=1
(∅ )
n
−1 i 2
∑ {exp −
. x ( i) n+1
}
i=1
Differentiate above equation with respect to ' ∅ ' and equating to zero we have non-linear equation which is given by
Johnson et al. (2013).
∅^ LSE =
−∑
n
exp
( )∑ ( )
−2
∅ . x ( i)
+
n
exp
−1
∅ . x (i )
.
i
=0
i=1 x (i) i=1 x (i ) n+1
We can use Newton- Raphson method to solve the non-linear equation [7].
1 Gamma Prior
The pdf of gamma prior is defined as:-
α
β α −1 − β ∅
γ 2 ( ∅ )= .∅ . e ; ∅ >0 , α , β >0 (7)
ᴦα
2 Exponential Prior
The pdf of Exponential prior is defined as:-
− ∅ .δ
γ 3 ( ∅ )=∅ . e ; ∅ >0 , δ >0(8)
So the posterior distribution corresponding to each prior as:-
w 1 ( ∅|x )=¿¿
w 2 ( ∅|x )=¿ ¿
Lh ( ∅^
h , ∅ ) =¿
∅ h=E w ( ∅ ∨x )
Corresponding to w 1 ( ∅|x ) :-
(β+ t)
∅^s 1= (12)
(n−α−1)
Corresponding to w 2 ( ∅|x ) :-
(δ+ t)
∅^s 2= (13)
(n−3)
3.5 Bayes Estimators under ELF
The Entropy loss function was proposed by James and stein. It is defined as :-
L ( ∅ , ∅ ELF ) =w
[( ) ( ) ]
∅ ELF
∅
−ln
∅ ELF
∅
−1 (14)
∅ ELF =¿
Now the bayes estimator for the parameter of ∅ under Entropy loss function corresponding to the three posterior
distributions are given below.
Corresponding to w 1 ( ∅|x ) :-
(β+t )
∅ E 1= (15)
(n−α )
Corresponding to w 2 ( ∅|x ) :-
→ ( 1∅|x ) G (n−2 , δ+ t)
Therefore,
( δ+t)
∅ E 2= (16)
(n−2)
3.6 Bayes Estimators under QLF
A new symmetric loss function named as Quadratic loss function(QLF) was defined as:
L ( ∅ , ∅QLF )=¿
Corresponding to w 1 ( ∅|x ) :-
Corresponding to w 2 ( ∅|x ) :-
→ ( 1∅|x ) G (n−2 , δ+ t)
Hence,
(δ+ t)
∅ Q 2= (19)
(n−1)
4- Simulation Analysis
A comprehensive simulation study conducted for different parameter points to highlight the performance of
estimators ∅^ MLE , ∅^ LSE , ∅^s 1 , ∅^s 2 , ∅ E 1 , ∅ E 2 , ∅Q 1 , ∅ Q 2 using Mean square error (MSE) and Mean absolute
percentage error (MAPE) procedure under different sample size.
MSE(∅^ ¿=E ¿
MLE LSE
From table 1, it is noted that if the sample size is small then the values of LSE Perform better as it is close to the true
value of the parameter. But when sample size increases the values of MLE performs better than LSE as it become
closer to the true value. Also the MSE and MAPE both are shows decreasing behavior as the sample size increases.
Table 2. Values For the Bayes Estimator of ∅ Under Gamma Prior Using
( α =5 , β=5 ) , ( α =4 , β=3 ) ,(α =6 , β=10)
Table 3. Values for the Bayes Estimator of ∅ Under Exponential Prior Using ∅ =1 , 1.5 , 3
From table 3, Corresponding to the values of different loss functions under exponential prior shows the overestimate
values of Bayes estimate under all sample size. Distinct sample sizes and parameter values are allocated. Under keen
observation it is noted that Entropy loss function (ELF) performs better than all other loss functions.
5- Conclusion
From the above analysis it is concluded that the Bayesian approach performs better than the classical approach as
their Mean square error (MSE) and Mean absolute percentage error (MAPE) shows much consistency.
References
1. Lemonte, A. J. 2013. A new exponential-type distribution with constant, decreasing, increasing, upside-down
bathtub and bathtub-shaped failure rate function. In: Computational Statistics Data Analysis 62: 149-170.
2. Singh, S. K., U. Singh and D. Kumar. 2013. Bayes estimators of the reliability function and parameter of inverted
exponential distribution using informative and non-informative priors. In: Journal of statistical computation and
simulation 83: 2258-2269.
3. Lin, C. T., B. S. Duran and T. O. Lewis. 1989. Inverted gamma as a life distribution. Microelectronics reliability
29: 619-626.
4. Al-Noor, N. H. and S. F. Bawi. 2015. Bayes estimators for the parameter of the inverted exponential distribution
under symmetric and asymmetric loss functions. In: Journal of natural sciences research 5: 45-42.
5. Krings, A. W. and Z. Ma. 2008. Competing risks analysis of reliability, survivability, and prognostics and health
management.
6. Tiwari, R. C., Y. Li and M. S. Goodman. 2006. Survival analysis with change point hazard functions. In: Harvard
university biostatistics working paper series.
7. Johnson, R.W., D. Kliche and P.L. Smith. 2013. Maximum likelihood estimation of gamma parameters for
coarsely binned and truncated raindrop size data. In: Journal of Royal meteorological society 140:1245-1256.