The Generalized Unit Weibull Distribution: Properties, Estimation, and

Applications in Actuarial Science and Insurance

Mintodê Nicodème Atchadéa,1,, Géoffroy Bill Zoffouna , Mahoulé Jude Bogninoua , Théophile Otodjia , Aliou Moussa
Djibrila
a National Higher School of Mathematical Engineering and Modeling, National University of Sciences, Technologies, Engineering and

Mathematics, Abomey, Benin Republic

b University of Abomey-Calavi/ International Chair in Mathematical Physics and Applications (ICMPA: UNESCO-Chair), 072 BP 50 Cotonou, Rep.

Benin

Abstract
This study introduces the Generalized Unit Weibull (GUW) distribution, an extension of the Unit Weibull distri-
bution achieved through transformation and the inclusion of additional parameters. We explore key theoretical
properties of this novel distribution, including stochastic functions, quantile functions and measures, moments,
and Rényi entropy. The model’s unknown parameters are estimated using the maximum likelihood method. To
demonstrate its applicability, we compare the proposed model with existing alternatives using two real-world data
sets, particularly in actuarial science and insurance.
Keywords: Generalized Weibull distribution; Moments; Rényi Entropy; Statistical properties; Quantile regression

1. Introduction distribution[9], unit power Weibull distribution[10],

unit Xgamma Distribution[11], unit exponenti-
Statistical distributions are extensively used in numer- ated Fréchet distribution[12], transmuted Marshall-
ous fields, offering precious tools for decision-making. Olkin extended Topp-Leone distribution[13], unit-
They are used in life cycle analysis, system trusta- exponentiated half-logistic distribution[14], new mod-
bility, life expectation determination, insurance opin- ified kumaraswamy distribution[15], unit Burr XII[16],
ions, engineering, finance, economics, biology, ex- extension of J-shaped distribution[17], unit-Chen
treme event threat assessment, drug, husbandry, ac- distribution[18], new regression model for bounded
tuarial modeling, demography, administration, sports, response variable[19], unit generalized log Burr XII
and accouterments wisdom. distribution[20], unit-Rayleigh distribution[21], unit-
Among the new statistical distributions proposed re- Weibull distribution[22], new power Topp-leone
cently, those whose domain is bounded by the in- distribution[23], power Topp–Leone exponential neg-
terval [0, 1] are of particular interest because of their ative family of distributions[24], Topp-Leone Cauchy
suitability for representing empirical data within this family of distributions[25], two-parameter family
range, such as quotients, ratios, or percentages. This of distributions[26], Topp-Leone Cauchy Family of
type of quantitative data is frequently encountered in distributions,[27].
various fields of study, such as hazard assessment,
psychology, economics, medical applications, and In addition to these earlier findings, recent devel-
engineering. Distributions whose probability den- opments in statistical distributions have introduced
sity functions can adopt specific shapes, such as in- promising new concepts. However, numerous statisti-
creasing, decreasing, or bathtub, are particularly valu- cal distributions are limited in adapting to various data
able for modeling complex phenomena. Among the sets.
new distributions recently introduced are the fol- Certainly, some datasets show distinct features such as
lows: unit exponential distribution[1], unit upper trun- high skewness, kurtosis, heavy tails, inverted J-shapes,
cated Weibull distribution[2], unit Gumbel type-II multimodality, etc. Distribution generators offer the
distribution[3], generalized unit half-logistic geomet- possibility of efficiently managing, and manipulating
ric distribution[4], unit inverse exponentiated Weibull these dataset characteristics. We aim to develop a new
distribution[5], Unit modified Burr-III distribution[6], distribution in this study by generalizing the Weibull
kumaraswamy unit-Gompertz distribution[7], unit distribution and making it unitary. This transforma-
Muth distribution[8], marshall–Olkin reduced Kies tion is motivated by the need for greater flexibility in

∗ Corresponding author

Email address: [email protected] (Mintodê Nicodème Atchadé)

2 GENERALIZED UNIT WEIBULL DISTRIBUTION

modeling bounded data on the unit interval, particu- Where the distribution’s shape parameter is k > 0, the
larly in fields such as reliability analysis, survival mod- scale parameter is λ > 0.
1
eling, and proportions data. We demonstrate the high Using the transformation y = t α (t = y α ) and then
y xβ
degree of adaptability of the distribution to real-world the transformation x = y+β (y = 1−x ), we have a new
data using two applications: materials engineering and generalized distribution on (0, 1), that we call the GUW
finance. Weibull distribution is widely used because distribution. Given is its CDF by:
of its advantageous attributes, such as its probabilistic  βx
α k
function’s mathematical simplicity and flexibility. (1−x)
− λ
The article’s remaining sections are organized as fol- F (x, α, β, λ, k) = 1 − e , x ∈ ]0; 1[ (1)
lows: Section (2) presents a description of the Gen-
eralized Unit Weibull (GUW) distribution. Section (3) The hazard rate function (hrf) and related PDF are pro-
addresses some noteworthy characteristics. Sections vided by:
(4) and (5) provide the methodology for actuarial mea-

α k
sures and distribution parameters estimation. Sections
 βx
αk−1 (1−x)
 −
(6), (7), and (8) are devoted to the simulations, applica- αk β βx λ
f (x, α, β, λ, k) = e ,
tions, and new quantile regression model, in that order. (λ)k (1 − x)2 1−x
Finally, the conclusion is made in the section (9). (2)
and  αk−1
αk β βx
2. Generalized Unit Weibull distribution hrf = ,
(λ)k (1 − x)2 1−x
We propose a new generalized distribution with sup- where α, β, λ, k > 0.
port on the unit-interval (0, 1), which arises from a Figures (1), (2), and (3) show the CDF, PDF, and HRF
certain transformation on the two-parameter Weibull of the GUW distribution, respectively. Figure (1) illus-
distribution[28] with probability distribution function trates the flexibility of the cumulative function across
in (PDF) : different parameter settings. Figure (2) shows that the
 k−1 PDF can take various shapes, including decreasing, re-
k t t k versed J, or asymmetric. Figure (3) highlights the wide
h(t, λ, k) = e−( λ ) , t > 0, range of possible hazard rate behaviors, such as in-
λ λ
creasing, decreasing, or bell-shaped. This observation
and cumulative distribution function (CDF) : is consistent with prior findings in the literature. These
t k
curvature characteristics are widely understood and
H(t, λ, k) = 1 − e−( λ ) . important for developing universal statistical models.

Figure 1: CDF of the GUW distribution

2
3 SOME MATHEMATICAL FEATURES OF THE GUW DISTRIBUTION

Figure 2: PDF of the GUW distribution

Figure 3: HRF of the GUW distribution

3. Some mathematical features of the GUW distri- and

bution U (x, σ) = xk+j (6)

This part focuses on numerous relevant mathematical

characteristics of the GUW distribution.
Proof:
According to ( 1),
3.1. Series development of the density function f

α k
Proposition 1:
 βx
(1−x)
− λ
F (x, α, β, λ, k) = 1 − e , x ∈ ]0; 1[
The series development of f is provided by:
 βx

α k
∞ X
∞ αmk−1 ∞ (1−x)
X X X Considering G(x,σ) = λ
f (x, σ) = Tn · U (x, σ), (3)
m=0 k=0 i=0 j=0

where
     mk F (x, σ) = 1 − e−[G(x,σ)]
k+1 i+j−1 αmk − 1 1
Tn = (4)
k j i λ
zm
P+∞
m+αmk+i knowing that : ez = m=0 m!
[−1]
× αmkβ αmk , (5)
m! So,
3
3 SOME MATHEMATICAL FEATURES OF THE GUW DISTRIBUTION 3.2 Rényi Entropy

+∞ m
X [−G(x, σ)] 3.2. Rényi Entropy
F (x, σ) = 1 −
m=0
m!
The Rényi entropy for the distribution is defined as:
+∞ m
X [−1] m
=1− [G(x, σ)] ( +∞ )
m! 1 X ′
m=0 ER(X) = log Tn · Iγ (x, σ) ,
1−γ n=0
m
Let’s develop [G(x, σ)] ,
where
 α mk  nk+kγ
βx n
′ [−γ] 1 γ
[G(x, σ)]
m
=
(1−x) Tn = β αnk+γαk (αk) ,
n! λ

λ
 mk  αmk and
1 x
= β αmk , γ  γαk−γ+αnk
λ 1−x
Z 
1 x
Iγ (x, σ) = dx
R (1 − x)2 1−x

+∞ m  mk
X [−1] 1
F (x, σ) = 1 − (7) Proof:
m=0
m! λ
αmk
The Rényi entropy of X in the case of a continuous ran-

x
× β αmk , (8) dom variable is defined by :
1−x
Z 
1
ER(X) = log f (x, σ)γ dx , γ ̸= 1, γ sup 1
By differentiating expression (7) with respect to x, we 1−γ R
obtain the series expansion of f (x).

+∞
X [−1]
m  mk
1 αmk Considering (2)
f (x, σ) = − β αmk 2
m=0
m! λ (1 − x)  kγ  γ  γαk−γ
γ γ 1 1 βx

x
αmk−1 (f (x, σ)) = (αk) βγ
× λ (1 − x)2 1−x
1−x  βx
α k
(1−x)
−γ λ
∞ 
×e .

1 X k+1 k
2 = x ,
(1 − x) k
k=0

γ
kλ λ
The terms: (αk) , λ1 , β being constant then their
More explicitly, series developments remain unchanged.
 αmk−1 αmk−1 ∞  
x αmk−1
X X i+j−1
= (−1)
1−x i=0 j=0
j  βx

α k
(1−x)
−γ +∞ n  nk  αnk

αmk − 1
 λ X [−γ] 1 x
× (−1)i xj e = β αnk
i n=0
n! λ 1−x

Finally the development in series of f (x) gives :

∞ X
X ∞ αmk−1
X X ∞ so
f (x, σ) = Tn · U (x, σ),
+∞ n  nk  kγ
m=0 k=0 i=0 j=0 γ
X [−γ] 1 γ 1
(f (x, σ)) = β αnk (αk)
n=0
n! λ λ
where
 γ  γαk−γ  αnk
  
k + 1 i + j − 1 αmk − 1
  mk
1 γ 1 βx x
Tn = ×β
k j i λ (1 − x)2 1−x 1−x
m+αmk+i +∞
X [−γ]n 1   nk+kγ
[−1] = β αnk+γ+γαk−γ
× · αmkβ αmk , n! λ
m! n=0
 γ  γαk−γ+αnk
and γ 1 x
× (αk)
U (x, σ) = xk+j , (1 − x)2 1−x

4
3 SOME MATHEMATICAL FEATURES OF THE GUW DISTRIBUTION 3.3 Moments and associated measures

γ
Thus the series expansion of (f (x, σ)) gives :
+∞ n  nk+kγ  γ  γαk−γ+αnk
γ
X [−γ] 1 αnk+γαk γ 1 x
(f (x, σ)) = β (αk) . (9)
n=0
n! λ (1 − x)2 1−x

Replacing (9) in (3.2) gives :

(Z +∞ γαk−γ+αnk )
1 X [−γ]n  1 nk+kγ γ

1
γ 
x
ER(X) = log β αnk+γαk (αk) dx .
1−γ R n=0 n! λ (1 − x)2 1−x

Renyi’s Entropy finally gives :

( +∞ γαk−γ+αnk )
1 X [−γ]n  1 nk+kγ γ
Z 
1
γ 
x
αnk+γαk
ER(X) = log β (αk) 2
dx .
1−γ n=0
n! λ R (1 − x) 1−x

Posing :
n  nk+kγ
′ [−γ] 1 γ
Tn = β αnk+γαk (αk) ,
n! λ
Z  γ  γαk−γ+αnk
1 x
Iγ (x, σ) = dx
R (1 − x)2 1−x

we have
( +∞ )
1 X ′
ER(X) = log Tn · Iγ (x, σ)
1−γ n=0

3.3. Moments and associated measures so Z 1

Ms = xs × f (x)dx.
At this stage, we’ll look more closely at the moments of 0
the new distribution. Momentum is a crucial statisti- Using the series development (3), we obtain:
cal concept that aids in understanding a distribution’s
characteristics and movement as well as its form. Z 1 ∞ X
X ∞ αmk−1
X X ∞
Ms = xs × Tn U (x, σ)dx
0 m=0 k=0 i=0 j=0
Proposition: ∞ X
∞ αmk−1 ∞ Z 1
X X X
The GUW distribution’s moment of order s may be cal- = Tn xk+j+s dx
culated as: m=0 k=0 i=0 j=0 0

∞ X
∞ αmk−1 ∞ ∞ X
∞ αmk−1 ∞
X X X Tn
X X X Tn
Ms = , (10) Ms =
s+j+k+1 m=0 k=0 i=0 j=0
s+j+k+1
m=0 k=0 i=0 j=0

where Tn is defined respectively in (4).

For values of r equal to 1, 2, 3, and 4, we successively get
Proof: the first four moments of the GUW distribution: mean,
variance, skewness, and kurtosis.
A variable’s moment of order s is determined as follows: The figures (4) show the mean, variance and (5) show
skewness, and kurtosis of the GUW model for different
Ms = E(X s ), parametric values of α, β, λ, and k.

5
3 SOME MATHEMATICAL FEATURES OF THE GUW DISTRIBUTION 3.4 Moment Generating Function

Figure 4: Mean and variance of the GUW model with λ = 1 and k = 0.09

Figure 5: Skewness and kurtosis of the GUW model with λ = 2 and k = 20

+∞
!
3.4. Moment Generating Function X (tX)r
Proposition : MX (t) = E
r=0
r!
The moment-generating function is sometimes also
+∞ 
(tX)r

called the characteristic function. It is used to fully de- X
= E
scribe the distribution of a random variable in terms of r!
r=0
its moments.
The characteristic function is defined as follows: +∞ r
X t
∞ X
+∞ X ∞ αmk−1 ∞ r MX (t) = E (X r ) . (11)
X X X t Tn r=0
r!
MX (t) = ,
r=0 m=0 k=0 i=0 j=0
r! r + j + k + 1
The moment of order r of the distribution is repre-
Proof: sented by E (X r ),

The moment-generating function is defined by: Mr = E (X r )

MX (t) = E eXt .


By replacing (10) in (11) we have:

∞ X
+∞ X ∞ αmk−1 ∞ r
Knowing that the development in a series of exponen- X X X t Tn
tial gives : MX (t) = ,
r=0 m=0 k=0 i=0 j=0
r! r + j + k + 1
+∞
X (tx)r
etx = . 3.5. Quantile function
r=0
r!
Proposition:
We can write : We define the quantile function which corresponds to
6
3 SOME MATHEMATICAL FEATURES OF THE GUW DISTRIBUTION 3.5 Quantile function

 1 1
  1 1

the distribution as follows: βx = λ α [− ln(1 − t)] αk − x λ α [− ln(1 − t)] αk .
 1 1

λ α [− ln(1 − p)] αk Let’s arrange terms containing x in a single member
Q(p, σ) = h  1 1
i .
β + λ α [− ln(1 − p)] αk  1   1 
1 1
βx + x λ α [− ln(1 − t)] αk = λ α [− ln(1 − t)] αk ,
Proof :

Posons πt = Q(t, σ) ∀ y ∈ [0, 1] . h  1 i  1 

1 1

The quantile function is defined as πt , which is the so- β + λ α [− ln(1 − t)] αk ·x = λ α [− ln(1 − t)] αk .
lution to the following nonlinear equation:

t = F (x, σ).
Knowing that α , λ, β are strictly greater than 0, then we
So, have:
 βx

α k  1 1

−
(1−x) λ α [− ln(1 − p)] αk
λ
1−t=e . π=h  1 1
i .
β + λ α [− ln(1 − p)] αk
Applying a log transformation to each member of the
equation, we obtain:
The UWG distribution’s 25%, 50%, and 75% quartiles

βx
α k may be found by adjusting p =0.25, p =0.5, and
(1−x) p =0.75, respectively, in equation (3.5).
− ln(1 − t) =   (12)
λ Assume that p is evenly distributed (0, 1), in this case,
the following random data sets of size n can be gener-
Let’s raise each number in the equation (12) to the ated by the QF using the GUW distribution:
power (1/k)  
1 1

α λ α [− ln(1 − yi )] αk
πi = h i , i = 1, 2, ...., n

1 βx  1 1
λ [− ln(1 − t)] k = . (13) β + λ α [− ln(1 − yi )] αk
(1 − x)

Let’s raise each number in the equation (13) to the Graphs of Bowley and Moor skewness and kurtosis are
power (1/α) shown in (6).

(a) Plot of the Bowley’s coefficient of skewness (b) Plot of the Moor’s coefficient of kurtosis

Figure 6: Skewness (left) and Kurtosis (right) plots

7
3 SOME MATHEMATICAL FEATURES OF THE GUW DISTRIBUTION 3.6 Survival function

 βx

α  k
3.6. Survival function (1−x)
− λ
The GUW distribution is characterized by its survival suf (x) = e .
function, which is as described below:

suf (x) = 1 − F (x),

Figure 7: Survuival functions of the GUW distribution

 αk−1
3.7. Hazard function αk β βx
haf (x) = .
The GUW distribution’s hazard function may be de- (λ)k (1 − x)2 1−x
scribed as follows:
f (x)
haf (x) = ,
suf (x)

Figure 8: haf of the GUW distribution

3.8. Cumulative hazard function (cf ) So, the cf of the GUW distribution is as follows:
The GUW distribution is characterized by its cf , which  α k
βx
is defined as follows : (1−x)
cf (x) =   .
cf (x) = − log(suf (x)), λ

8
3 SOME MATHEMATICAL FEATURES OF THE GUW DISTRIBUTION 3.9 Reserve hazard function(Rf)

Figure 9: Cf of the GUW distribution

 βx

α k
3.9. Reserve hazard function(Rf) (1−x)
− λ
Poses : A(x) = e .
The GUW distribution is characterized by its Rf, : so
 αk−1
f (x) αk β βx A(x)
Rf (x) = , Rf (x) =
F (x) (λ)k (1 − x)2 1−x 1 − A(x)

Figure 10: Rf of the GUW distribution

3.10. Average absolute deviation mean of µ, the average absolute deviation is calculated
The average absolute deviation indicates how far, on below:
average, each piece of data in a set is from the mean mad(µ) = E(|X − µ|) (14)
of that set. If we consider a GUW distribution with a

By using (14), we have :

9
4 ACTUARIAL MEASURES 3.11 Median absolute deviation (MD)

Z 1
mad(µ) = |x − µ|f (x) dx
0
Z µ Z 1
= (−x + µ)f (x) dx + (x − µ)f (x) dx
0 µ
∞ X
X ∞ αmk−1
X X ∞ Z µ Z 1 
= Tn (−xk+j+1 + µxk+j ) dx + (xk+j+1 − µxk+j ) dx
m=0 k=0 i=0 j=0 0 µ

∞ X
∞ αmk−1 ∞
1 − 2µk+j+2 2µk+j+2 − µ
X X X  
= Tn + .
m=0 k=0 i=0 j=0
k+j+2 k+j+1

So, the average absolute deviation is expressed as:

∞ X
X ∞ αmk−1
X X ∞
mad(µ) = Tn Mµ (σ),
m=0 k=0 i=0 j=0

where
1 − 2µk+j+2 2µk+j+2 − µ
Mµ (σ) = + .
k+j+2 k+j+1
3.11. Median absolute deviation (MD) have a GUW distribution with a median of me, the MD
may be stated as follows:
It’s comparable to the MAD, except instead of utilizing
the mean as a reference point, we use the median. If we M D(me) = E(|X − me|), (15)

By using (15), we have :

Z 1
MD(me) = |x − me| × f (x) dx
0
Z me Z 1
= (−x + me) × f (x) dx + (x − me) × f (x) dx
0 me
X ∞ αmk−1
∞ X X X ∞ Z me Z 1 
k+j+1 k+j k+j+1 k+j
= Tn (−x + me · x ) dx + (x − me · x ) dx
m=0 k=0 i=0 j=0 0 me

∞ X
∞ αmk−1 ∞
1 − 2mek+j+2 2mek+j+2 − me
X X X  
= Ti,j,k,n × + .
m=0 k=0 i=0 j=0
k+j+2 k+j+1

So, the MD is given by :

∞ X
X ∞ αmk−1
X X ∞
MD(me) = Tn Mme (x, σ),
m=0 k=0 i=0 j=0

where
1 − 2mek+j+2 2mek+j+2 − me
Mme (σ) = + .
k+j+2 k+j+1
4. Actuarial measures 4.0.1. VaR measure
The VaR of the GUW distribution is defined by :
 1

This section covers the theoretical and practical ele- λα [− ln(1 − q)] αk
ments of numerous essential risk measures. These in- VaRq = h  1
i .
clude Value at Risk (VaR), Average Loss Size over VaR β + λα [− ln(1 − q)] αk
(TVaR), VaR Size (TV), and VaR Probability (TVP) for
the new distribution. Proof. The VaR of a random variable is the quantile of
10
4 ACTUARIAL MEASURES

its distribution function, denoted by VaRq , and can be So we have :

expressed as follows:  1

λα [− ln(1 − q)] αk
VaRq = h  1
i .
VaRq = Q(q). β + λα [− ln(1 − q)] αk

Figure 11: The VaR plot

4.0.2. Mesure TVaR so:

The TVaR of the GUW distribution is described by : Z 1
1
∞ X
∞ αnk−1 ∞ TVaRq = xf (x) dx
1 X X X 1−q VaRq
TVaRq = Tn IV(σ),
1−q Z 1 ∞ X
∞ αmk−1 ∞
m=0 k=0 i=0 j=0 1 X X X
= x Tn U (x, σ)
where 1−q VaRq m=0 k=0 i=0 j=0
!
∞ ∞ αmk−1 ∞
1 − VaRk+j+2
q 1 XX X X
Z 1
IV(σ) = . = Tn xk+j+1
k+j+2 1 − q m=0 VaRq
i=0 j=0
k=0
∞ X
∞ αmk−1 ∞
!
Proof 1 X X X 1 − VaRk+j+2
q
= Tn .
TVaR is defined by: 1−q k+j+2
m=0 k=0 i=0 j=0
Z 1
1 so , TVaRq
TVaRq = xf (x) dx.
1−q VaRq
∞ ∞ αmk−1 ∞
!
1 XX X X 1 − VaRk+j+2
q
Knowing that f (x) gives : TVaRq = Tn .
1 − q m=0 i=0 j=0
k+j+2

α k k=0
βx
αk−1 (1−x)
 −
αk β βx λ
f (x, σ) = e
(λ)k (1 − x)2 1−x
∞ X
X ∞ αmk−1
X X ∞
= Tn U (x, σ)
m=0 k=0 i=0 j=0

11
4 ACTUARIAL MEASURES

Figure 12: The TVaR plot

4.0.3. TV measure Proof.

The GUW distribution TV is defined by : The TV distribution may be characterized as:

TVq (X) = E(X 2 |X > VaRq ) − (TVaRq )2

∞ X
∞ αmk−1 ∞ Z 1
1 X X X 1
TVq = Tn IV’(σ) − (TVaRq )2 , = x2 f (x) dx − (TVaRq )2
1−q m=0 k=0 i=0 j=0 1 − q VaRq
∞ ∞ αmk−1 ∞
1 XX X X
where = Tn
1 − q m=0
k=0 i=0 j=0
! !
1 − VaRqk+j+3 1 − VaRk+j+3
q
IV’(σ) = × − (TVaRq )2
k+j+3 k+j+3

Figure 13: The TV plot

4.0.4. TVP Measure

TVP is another key metric used in insurance science.
Distribution TVP has the following form:

TVPq (x) = TVaRq + λTVq

where
∞ ∞ αmk−1 ∞
1 XX X X
TVq = Tn IV’(σ) − (TVaRq )2
1 − q m=0 i=0 j=0
k=0

12
5 ESTIMATION

and
∞ ∞ αmk−1 ∞
1 XX X X
TVaRq = Tn IV(σ)
1 − q m=0 i=0 j=0
k=0

Figure 14: The TVp plot

5. Estimation

Let x1 , x2 , . . . , xm be a random sample of size m from the variable X. Employing the PDF provided in (2), the like-
lihood function may be expressed as follows:
m
Y
ℓ(α, β, λ, k) = f (xk ),
k=1

So, we have :
  
m m m m m m βx k
kα ln (1−xj )
  X
X X X X X βxj j
1
ℓ(α̂, β̂, λ̂, k̂) = ln (αk)−k ln(λ)− ln(β)−2 ln(1−xj )+(αk−1) ln − e
j=1 j=1 j=1 j=1 j=1
1 − xj j=1
λ

The logarithmic likelihood function may be stated in the form of:

h i
ℓ(α̂, β̂, λ̂, k̂) = ln L(α̂, β̂, λ̂, k̂)) ,

We obtain :
  α k 
βxj
(1−xj )

αk−1 −
  
m   λ 
X  αk β βx j 
ℓ(α̂, β̂, λ̂, k̂) = ln 
 (λ)k (1 − xj )2 1 − xj e 

j=1  
 

 α k
m m m m m   m βxj
X X X X X βxj X (1−xj )
= ln (αk) − k ln(λ) − ln(β) − 2 ln(1 − xj ) + (αk − 1) ln −  
j=1 j=1 j=1 j=1 j=1
1 − xj j=1
λ

Introducing the maximum likelihood estimators α̂, β̂, λ̂, and k̂.
We check:

13
6 A NEW QUANTILE REGRESSION MODEL

ℓ(α̂, β̂, λ̂, k̂) = max(α̂,β̂,λ̂,k̂)∈[0,+∞]4 l(α̂, β̂, λ̂, k̂)

The first partial derivatives of l(α̂, β̂, λ̂, k̂) with regard to zero are shown below:

  
m m βx k
kα ln 1−xj
  X  
∂l 1 X βxj βxj j 1
= +k ln + k ln e ,
∂α α j=1
1 − xj j=1
1 − xj λ

  x α k
m j
∂l 1 (αk − 1) X αk−1  (1−xj )
=− + − αk(β)  ,
∂β β β j=1
λ

    
m m βxj k m βx
kα ln (1−xj )
  X  
∂l 1 X βxj βxj kα ln 1 X
ln(λ)e−k ln(λ) ,
1−xj j
= − ln(λ) + (α) ln − α ln e + e
∂k k j=1
1 − xj j=1
1 − xj λ j=1
 
m βx
∂l k X kα ln (1−xjj )
=− + e kλ−k−1 .
∂λ λ j=1

Now that we have established the explicit formulas for To formulate the quantile regression model of the GUW
the partial derivatives concerning each parameter in distribution, we first proceed by making the parame-
the log-likelihood of the model, it becomes evident ter β the subject of the quantile function of the GUW
that solving the system analytically is not a feasible distribution. Then we replace it in CDF and PDF. Thus,
task. The complexity of the equations and the inter- after simplification, we obtain the cumulative distribu-
dependence of variables make it challenging to obtain tion function (QCDF) and quantile probability density
closed-form solutions. In such cases, the application (QPDF) of the GUW distribution.
of numerical methods becomes imperative. Numeri- Poses Q(p, σ) = µ
cal methods provide a practical and efficient approach  1 
1
to finding solutions, allowing us to navigate the intri- λ α [− ln(1 − p)] αk
cacies of the system and approximate solutions itera- µ= h  1 1
i
tively. β + λ α [− ln(1 − p)] αk
 1 1
  1 1

6. A new quantile regression model µβ + µ λ α [− ln(1 − p)] αk = λ α [− ln(1 − p)] αk
The concept of parametric quantile regression recently  1 1

gained popularity due to its robustness in modeling µβ = λ α [− ln(1 − p)] αk (1 − µ)
asymmetric data or data with extreme values. This type
of regression is also effective in dealing with asymmet- so  
1−µ
ric and high-tail response variables, which are defined β = G(p)
µ
on the interval (0,1). To implement these regressions, it
is necessary to re-parameterize the probability density where
functions (PDFs) of the distribution in terms of quan-  1 1

tiles, to get the quantile PDF ([29], [30], [31], [32], [4]). G(p) = λ α [− ln(1 − p)] αk

QCDF and QPDF are, respectively, given by

 α k
G(p)( 1−µ
µ )
y

 (1−y) 
− λ

 
QF (y, α, β, λ, µ, k, p) = 1 − e , y ∈ ]0; 1[
 1−µ
α k
G(p)(
µ )
y

 (1−y) 
    αk−1 − λ

1−µ
G(p) 1−µ
 
αk G(p) µ µ y
Qf (y, α, β, λ, k, µ, p) =   e y ∈ ]0; 1[
(λ)k (1 − y)2 1−y

14
6 A NEW QUANTILE REGRESSION MODEL

Where µ ∈ (0, 1) and p ∈ (0, 1). Figure (15), (16), (17), tilted, decreasing, increasing, symmetrical, J-shaped
and (18) shows respectively the plots of QCDFs and and bathtub-shaped. This shows that the regression
QPDFs for different quantiles and parameter values. model developed from this PDF is flexible enough to
QPDFs come in many shapes, including left- and right- deal with short-interval data with such properties.

(a) p=0.1 (b) p=0.25

Figure 15: QCDFs plots

(a) p = 0.50 (b) p = 0.75

Figure 16: QCDFs plots

15
6 A NEW QUANTILE REGRESSION MODEL 6.1 Estimation of regression parameters

(a) p = 0.1 (b) p = 0.25

Figure 17: QPDFs plots

(a) p = 0.50 (b) p = 0.75

Figure 18: QPDFs plots

When we have random observations y1 , y2 , y3 , ......., yn ing form:

from the GUW distribution, and non-random predic-  
tor variables x1 , x2 , x3 , ......., xn , the GUW quantile re- µi
logit(µi ) = log = xTi γ
gression is established by associating the conditional 1 − µi
quantile of the dependent variable with the predictor
variables through a suitable link function, as follows: 6.1. Estimation of regression parameters
g(µi ) = xTi γ To estimate the unidentified regression parameters us-
ing the ML technique, we compute the logarithmic
In this regression model, γ is the vector of coefficients likelihood by replacing:
for the independent variables, and xi represents the
vector of predictor variables for each observation i. The exp(xTi γ)
function g(−) is the connection function used. For this µi =
1 + exp(xTi γ)
study, the logit link function is preferred for its ease of
interpretation of the coefficients of the exogenous vari- in the quantile PDF. The log-likelihood function is so
ables. Thus, the regression structure takes the follow- defined as:

16
7 SIMULATION STUDY 6.2 Regression modeling for educational data

n n n n   n n
X X X X 1−µ X X
ℓ(α̂, β̂, λ̂, k̂, p̂) = ln(αk) − k ln(λ) + ln(G(p)) + ln −2 ln(1 − y) + (αk − 1) ln(G(p))+
i=1 i=1 i=1 i=1
µ i=1 i=1

(16)
 α k
G(p)·( 1−µ
µ )y
n   n n (1−y)
X 1−µ X X  
(αk − 1) ln + (αk − 1) ln(y) − (αk − 1) ln(1 − y) −  
i=1
µ i=1 i=1
 λ 

(17)
To get parameter estimates, we set the components of (measured by the mean score of the Cantril scale, also
the score vector to zero while concurrently solving the known as the Self-Anchoring Striving Scale, x2 ).
resultant system of equations. To fit the median regres- The regression model is formulated as follows:
sion, we put p=0.50 in equation (17) and maximize the logit(µi ) = γ0 + γ1 x1 + γ2 x2
log-likelihood function. The parameter standard error where µi represents the median for the GUW and
estimates are computed using the ML method’s large- GUHLG distributions. We calculate maximum like-
sample characteristic. As per [33], the Fisher informa- lihood estimates (MLE), associated standard errors,
tion matrix for parameter standard error estimation is: and estimated log-likelihood values for all models, as
shown in table (1). This reveals that only the coeffi-
∂ 2 ℓ(η|y) cients γ0 and γ1 of the GUW model are significant at the
I(η̂) = −
∂η T ∂η η=η̂ 0.05 threshold, while for the GUHLG model, none of
the coefficients is significant. We also observe a nega-
6.2. Regression modeling for educational data tive relationship between the level of education (repre-
In this section, we carry out an analysis of the real sented by percentage) and the country’s homicide rate,
data to compare the new regression with the Gener- but a positive relationship between the level of edu-
alized Unit Half-Logistic Geometric (GUHLG) regres- cation and satisfaction with life. These results suggest
sion model. The data can be accessed via the link that an increase in life satisfaction is associated with an
https://stats.oecd.org/index.aspx?DataSetCode=BLI increase in the percentage of educational achievement,
[32]. They include three variables: level of education while an increase in the homicide rate corresponds to
(expressed as a percentage of the 35 OECD countries, a decrease in the percentage of educational achieve-
y), homicide rate (as a ratio, x1 ), and life satisfaction ment.

Table 1: The result for fitted regression models.

GUW GUHLG
Estimate SE p-value Estimate SE p-value
γ0 12.8942 6.3630 0.0427∗ -1.720182 4.454259 0.699
γ1 -2.3384 1.0181 0.0216∗ 0.010572 0.653999 0.987
γ2 0.2337 0.1547 0.1309 -0.001193 0.076603 0.988

7. Simulation study rameters of concern. Next, we evaluate two important

statistical metrics that show how reliable and accurate
In this section, we check whether the estimates ob-
the estimations are: mean bias (Bias) and root mean
tained by the maximum likelihood method (MLE) are
square error (RMSE). The outcomes of this investiga-
consistent for the GUW distribution. To do this, we
tion are shown in table (2). This simulation technique
carry out a simulation study using R software. We cre-
improves our understanding of MSE performance and
ate a thousand independent samples of different sizes
consistency across multiple sample sizes for the GUW
(100, 200, 300, 400, and 500) from the GUW distribu-
distribution.
tion. For each sample, we compute the MLEs of the pa-

17
8 DATA HANDLING

Table 2: Values for mean, mean bias, and RMSE of simulations for GUW distribution.

Sample size α = 0.5 β = 2 λ = 5 k = 0.8 α = 0.6 β = 1.5 λ = 3 k = 5

n Mean RMSE Bias Mean RMSE Bias
100 0.3865726 0.3403587 0.003719803 0.5160878 0.09829673 -0.0007436222
200 0.3871267 0.3338191 0.002655435 0.5160524 0.09050732 -0.0060478341
α 300 0.3875297 0.3378054 0.006666082 0.5161166 0.09409248 -0.0024559742
400 0.3877250 0.3397468 0.007578518 0.5161333 0.09463990 -0.0026778599
500 0.3875603 0.3411975 0.010500618 0.5161374 0.09624244 -0.0009811381

100 0.4703165 0.3581328 0.003394391 0.7537197 0.08557860 -0.0006667927

200 0.4708478 0.3486794 -0.001702143 0.7536183 0.07560987 -0.0061186074
β 300 0.4711307 0.3545124 0.004974935 0.7536426 0.07831139 -0.0029596014
400 0.4713905 0.3554896 0.006533760 0.7536522 0.07828130 -0.0030667419
500 0.4712644 0.3582302 0.011011426 0.7536705 0.08027758 -0.0014813712

100 0.2917571 0.3026817 0.003763988 0.3319138 0.08227289 -0.0003214238

200 0.2922474 0.3003471 0.004936670 0.3319190 0.07712006 -0.0041476335
λ 300 0.2927800 0.3025652 0.006881418 0.3319927 0.07986454 -0.0013732919
400 0.2928630 0.3055884 0.006582468 0.3320070 0.08059554 -0.0016638122
500 0.2926784 0.3062275 0.008259311 0.3320030 0.08181845 -0.0003755340

100 0.3932002 0.3125611 0.0031226059 0.4643264 0.1986186 -0.0006312249

200 0.3936681 0.3055806 -0.0001105596 0.4644523 0.1885544 -0.0086263304
300 0.3940392 0.3101794 0.0048216971 0.4646319 0.1952523 -0.0020652042
k 400 0.3942010 0.3120068 0.0052970049 0.4646903 0.1962480 -0.0024236803
500 0.3940626 0.3139784 0.0085886883 0.4646540 0.1987249 0.0008041739

8. Data handling property is essential in many applications such as

the manufacture of composite materials used in the
We will utilize two appropriate data sets and compare
aerospace, automotive, and construction industries.
them to the current competitive models listed below to
The National Physical Laboratory (NPL) is the UK’s na-
assess the efficacy of the distribution in practical set-
tional metrology laboratory. They carry out precise
tings:
measurements in a variety of fields, including mate-
1. Exponentiated Topp-Leone distribution rials characterization. In this case, they measure the
(ETPLD)[34]: strength of glass fibers.
α−1
Data on the strength of 1.5 cm glass fibers measured
f (x; α, β) = 2αβ(1 − x) (x(2 − x)) at the National Physical Laboratory in England rep-
β−1 resent important information for understanding and
(1 − (xα (2 − x)α ))
characterizing the mechanical properties of glass
2. Kumaraswamy distribution (KwD)[35]: fibers, with potential implications in various indus-
trial and research fields. The data are presented below
f (x; α, β) = αβxα−1 (1 − xα )β−1 (https://catalog.data.gov/dataset/unemployment-
insurance-data-july-2008-to-april-2013) : 0.055,
3. Rayleigh distribution(RD)[36].: 0.093, 0.125, 0.136, 0.149, 0.152, 0.158, 0.161, 0.164,
0.168, 0.173, 0.181, 0.200, 0.074, 0.104, 0.127, 0.139,
x2
 
x 0.149, 0.153, 0.159,0.161, 0.166, 0.168, 0.176, 0.182,
f (x; α) = 2 exp − 2
α 2α 0.201, 0.077, 0.111, 0.128, 0.142,0.150, 0.154, 0.160,
0.162, 0.166, 0.169, 0.176, 0.184, 0.224, 0.081,0.113,
The predicted values of the PDF parameters were cal- 0.129, 0.148, 0.150, 0.155, 0.161, 0.162, 0.166, 0.170,
culated and visualized using R, Matlab online and 0.177,0.184, 0.084, 0.124, 0.130, 0.148, 0.151, 0.155,
Python under Spyder. Examination of the data allows 0.161, 0.163, 0.167, 0.170, 0.178, 0.189. Table (3) shows
us to check whether the distribution behaves as ex- the predicted values for dataset I, with initial values of
pected in real-life circumstances. α = 2, β = 6, λ = 8, and k = 5. The information
criteria obtained by various models for this dataset are
8.1. Dataset I summarized in table (4). In addition, the empirical
Glass fiber strength is a measure of a glass fiber’s abil- (PDFs) for dataset I are visible in the figure (19).
ity to resist breakage or deformation under stress. This
18
8 DATA HANDLING 8.1 Dataset I

Table 3: Predicted values for various models

Model α β λ k

UWG 2.042932 6.166541 7.907704 4.632182

ETPLD 1.857044 6.317736 - -

Kumaraswamy 1.656934 5.580659 - -

Rayleigh 1.597854 - - -

Table 4: Measures of selection for various models

Model AIC CAIC BIC HQIC

UWG -66.57406 -65.8844 -58.00152 -63.20243

ETPLD 82.25297 82.45297 86.53924 83.93878

Kumaraswamy -41.42698 -41.22698 -37.14071 -39.74117

Rayleigh 122.5984 122.6639 124.7415 123.4413

Figure 19: Visualization of PDFs for various models

19
8 DATA HANDLING 8.2 Dataset II

8.2. Dataset II programs aimed at them. It can also help identify

58 observations, or monthly unemployment insurance seasonal, economic, or geographic trends that influ-
measures, were made between July 2008 and April ence veterans’ claims. The data are presented below
2013. The State of Maryland, USA’s Department of La- ([37]):0.129, 0.103, 0.129, 0.125, 0.103, 0.111, 0.149,
bor, Licensing and Regulation, provided these statis- 0.115, 0.131, 0.106, 0.102, 0.138, 0.141, 0.140, 0.155,
tics. 0.149, 0.106, 0.132, 0.137, 0.118, 0.136, 0.157, 0.124,
Variable number 5, entitled “New.Claims.Filed...UCX”, 0.177, 0.170, 0.203, 0.184, 0.173, 0.153, 0.153, 0.166,
represents the number of new claims filed, 0.145, 0.145, 0.148, 0.144, 0.164, 0.166, 0.178, 0.171,
This variable, therefore, measures the monthly fre- 0.179, 0.166, 0.127, 0.207, 0.168, 0.192, 0.182, 0.193,
quency with which new claims are filed by veterans for 0.191, 0.195, 0.194, 0.156, 0.267, 0.180, 0.145, 0.207,
unemployment insurance benefits. It makes it possi- 0.159, 0.149, 0.172
ble to monitor unemployment trends among veterans
and the evolution of unemployment insurance claims Table (5) shows the estimated values for dataset II, with
in this specific demographic group. initial values of α =8, β =2, λ =5, and k =3. The in-
Analysis of this variable can be crucial in understand- formation criteria obtained by various models for this
ing the financial support needs of veterans, and in as- dataset are listed in table (6). In addition, the figure (20)
sessing the effectiveness of unemployment insurance presents the empirical probability (PDFs) for dataset II.

Table 5: Predicted values for various models

Model α β λ k

UWG 7.902857 2.305981 4.927164 2.806308

ETPLD 7.651511 2.087676 - -

Kumaraswamy 7.652575 2.151124 - -

Rayleigh 7.613136 - - -

Table 6: Measures of selection for various models

Model AIC CAIC BIC HQIC

UWG -57.56362 -56.8089 -49.32184 -54.35328

ETPLD 599.1901 599.4082 603.3109 600.7952

Kumaraswamy 1121.127 1121.345 1125.248 1122.732

Rayleigh 692.6526 692.724 694.713 693.4551

20
10 FUTURE WORK

Figure 20: Visualization of PDFs for various models

After examining tables (3), (4), (5), (6), as well as figures (19) and (20), we can conclude that the GUW model proves
to be more compatible with datasets I and II than competing models. Its flexibility enables it to adapt to domains as
varied as materials engineering and unemployment insurance data.

9. Conclusion as specific statistical tests or cross-validation tech-

niques. This would enable us to deepen our under-
In this study, we introduced and examined the GUW, a standing of the performance of these distributions in
novel statistical distribution. The Weibull distribution different contexts.
has been modified to produce this distribution. In addition, we could seek to improve the flexibility and
The order statistic, quantile power series, Renyi en- accuracy of the GUW distribution by introducing ad-
tropy, and moments were among the statistical charac- ditional parameters or exploring alternative distribu-
teristics of the distribution that were determined. We tion functions. This could enable us better to model
used the maximum likelihood estimation approach to complex phenomena or data with particular character-
build estimators for the distributions’ parameters. Ac- istics.
tuarial metrics have also been devised, including TVAR, Finally, we could explore applying the data simulation
VAR, TV, and TVP. approach in specific fields such as health, the envi-
Through the use of actual data and a performance ronment, or the social sciences. By adapting the pro-
comparison with other models now in use, such ET- posed methodology to the needs and particularities of
PLD, KwD, and RD, the applications of the GUW dis- these fields, we could contribute to concrete problem-
tribution in banking and the materials sector have also solving and informed decision-making.
been proven. The GUW quantile function was also By combining these approaches, we could extend the
used to create a quantile regression model. The find- potential applications of this work and provide valu-
ings indicate that professionals working in the fields of able tools for research and practice in various fields.
materials engineering and finance who perform statis-
tical analysis may find the GUW distribution useful.
Funding

10. Future work No funding.

We envisage several promising avenues of research. Availability of data and materials

Firstly, we could extend this study by exploring other
methods of validating customized distributions, such Data are provided within the paper.
21
REFERENCES REFERENCES

Competing interests [18] M. C. Korkmaz, E. Altun, C. Chesneau, H. M. Yousof, On the unit-

chen distribution with associated quantile regression and appli-
The authors declare they have no competing interest. cations, Mathematica Slovaca 72 (2022) 765–786.
[19] E. Altun, M. El-Morshedy, M. Eliwa, A new regression model for
bounded response variable: An alternative to the beta and unit-
lindley regression models, Plos one 16 (2021) e0245627.
References [20] F. A. Bhatti, A. Ali, G. Hamedani, M. Ç. Korkmaz, M. Ahmad, The
unit generalized log burr xii distribution: Properties and appli-
[1] H. S. Bakouch, T. Hussain, M. Tošić, V. S. Stojanović, N. Qar- cation, AIMS Mathematics (2021).
malah, Unit exponential probability distribution: Characteri- [21] R. A. Bantan, C. Chesneau, F. Jamal, M. Elgarhy, M. H. Tahir,
zation and applications in environmental and engineering data A. Ali, M. Zubair, S. Anam, Some new facts about the unit-
modeling, Mathematics 11 (2023) 4207. rayleigh distribution with applications, Mathematics 8 (2020)
[2] I. E. Okorie, E. Afuecheta, H. S. Bakouch, Unit upper truncated 1954.
weibull distribution with extension to 0 and 1 inflated model– [22] J. Mazucheli, A. Menezes, L. Fernandes, R. De Oliveira, M. Ghi-
theory and applications, Heliyon 9 (2023). tany, The unit-weibull distribution as an alternative to the ku-
[3] A. Shafiq, M. F. Baig, Y. Fatima, A. Haer, S. Parveen, M. A. Tariq, maraswamy distribution for the modeling of quantiles condi-
M. K. A. Khan, et al., Teratogenic effects of lead in the developing tional on covariates, Journal of Applied Statistics 47 (2020) 954–
chick embryos: Teratogenic effects of lead in the developing chick 974.
embryos, MARKHOR (The Journal of Zoology) (2023) 16–20. [23] M. N. Atchadé, M. N’bouké, A. M. Djibril, S. Shahzadi, E. Hus-
[4] S. Nasiru, C. Chesneau, A. G. Abubakari, I. D. Angbing, General- sam, R. Aldallal, H. M. Alshanbari, A. M. Gemeay, A.-A. H. El-
ized unit half-logistic geometric distribution: properties and re- Bagoury, A new power topp–leone distribution with applications
gression with applications to insurance, Analytics 2 (2023) 438– to engineering and industry data, PLoS one 18 (2023) e0278225.
462. [24] M. N. Atchadé, T. Otodji, A. M. Djibril, M. N’bouké, Power
[5] A. S. Hassan, R. S. Alharbi, Different estimation methods for topp–leone exponential negative family of distributions with nu-
the unit inverse exponentiated weibull distribution, Communi- merical illustrations to engineering and biological data, Open
cations for Statistical Applications and Methods 30 (2023) 191– Physics 21 (2023) 20230151.
213. [25] M. N. Atchadé, M. J. Bogninou, A. M. Djibril, M. N’bouké, Topp-
[6] M. A. u. Haq, S. Hashmi, K. Aidi, P. L. Ramos, F. Louzada, leone cauchy family of distributions with applications in indus-
Unit modified burr-iii distribution: Estimation, characteriza- trial engineering, Journal of Statistical Theory and Applications
tions and validation test, Annals of Data Science (2020) 1–26. 22 (2023) 339–365.
[7] I. U. Akata, F. C. Opone, F. E. Osagiede, The kumaraswamy unit- [26] M. C. Belili, A. M. Alshangiti, A. M. Gemeay, H. Zeghdoudi,
gompertz distribution and its application to lifetime datasets, K. Karakaya, M. Bakr, O. S. Balogun, M. N. Atchadé, E. Hussam,
Earthline Journal of Mathematical Sciences 11 (2023) 1–22. Two-parameter family of distributions: Properties, estimation,
[8] R. Maya, P. Jodra, M. Irshad, A. Krishna, The unit muth distribu- and applications, AIP Advances 13 (2023).
tion: statistical properties and applications, Ricerche di Matem- [27] M. N. Atchadé, M. J. Bogninou, A. M. Djibril, M. N’bouké, Cor-
atica (2022) 1–24. rection: Topp-leone cauchy family of distributions with applica-
[9] A. Z. Afify, M. Nassar, D. Kumar, G. M. Cordeiro, A new unit distri- tions in industrial engineering (2023).
bution: Properties, inference, and applications, Electronic Jour- [28] C. Lai, Distributions weibull, Revues interdisciplinaires Wiley :
nal of Applied Statistical Analysis 15 (2022) 438–462. Statistiques informatiques 3 (2011) 282–287.
[10] R. A. Bantan, S. Shafiq, M. Tahir, A. Elhassanein, F. Jamal, W. Al- [29] J. e. A. Mazucheli, Modèles de régression quantile et moyenne va-
mutiry, M. Elgarhy, et al., Statistical analysis of covid-19 data: sicek pour données limitées : nouvelle formulation, dérivations
Using a new univariate and bivariate statistical model, Journal mathématiques et applications numériques, Mathématiques 10
of Function Spaces 2022 (2022). (2022) 1389.
[11] S. Hashmi, M. Ahsan-ul Haq, J. Zafar, M. A. Khaleel, Unit [30] Mazucheli, Le modèle de régression quantile semi-normal
xgamma distribution: Its properties, estimation and applica- généralisé unitaire : formulation, estimation, diagnostics et ap-
tion: Unit-xgamma distribution, Proceedings of the Pakistan plications numériques, Informatique douce 27 (2023) 279–295.
Academy of Sciences: A. Physical and Computational Sciences [31] S. e. A. Nasiru, Nouvelle distribution de durée de vie pour la mod-
59 (2022) 15–28. élisation des données sur l’intervalle unitaire : propriétés, appli-
[12] A. G. Abubakari, A. Luguterah, S. Nasiru, Unit exponentiated cations et régression quantile, Applications mathématiques et
fréchet distribution: Actuarial measures, quantile regression and informatiques 27 (2022) 105.
applications, Journal of the Indian Society for Probability and [32] A. Gemeay, Distribution uniforme de power lambert : propriétés
Statistics 23 (2022) 387–424. statistiques, mesures actuarielles, analyse de régression et appli-
[13] F. C. Opone, J. E. Osemwenkhae, The transmuted marshall-olkin cations, AIP Avancées 13 (2023).
extended topp-leone distribution, Earthline Journal of Mathe- [33] B. G. Lindsay, B. Li, On second-order optimality of the observed
matical Sciences 9 (2022) 179–199. fisher information, The Annals of Statistics 25 (1997) 2172–2199.
[14] A. S. Hassan, A. Fayomi, A. Algarni, E. M. Almetwally, Bayesian [34] A. e. M. Pourdarvish, La distribution exponentiée de topp-leone
and non-bayesian inference for unit-exponentiated half-logistic : Propriétés et application, Journal des sciences appliquées de
distribution with data analysis, Applied Sciences 12 (2022) l’environnement et de la biologie 5 (????) 251–256.
11253. [35] P. Kumaraswamy, A generalized probability density function
[15] A. R. Alanzi, M. Q. Rafique, M. Tahir, W. Sami, F. Jamal, et al., A for double-bounded random processes, Journal of hydrology 46
new modified kumaraswamy distribution: actuarial measures (1980) 79–88.
and applications, Journal of Mathematics 2022 (2022). [36] Distribution de rayleigh et ses généralisations, Radio Science
[16] T. F. Ribeiro, F. A. Peña-Ramírez, R. R. Guerra, G. M. Cordeiro, An- Journal of Research NBS/USNC-URSI 68 (1964) 927–932.
other unit burr xii quantile regression model based on the differ- [37] A. Z. Afify, A. M. Gemeay, N. A. Ibrahim, The heavy-tailed expo-
ent reparameterization applied to dropout in brazilian under- nential distribution: risk measures, estimation, and application
graduate courses, Plos one 17 (2022) e0276695. to actuarial data, Mathematics 8 (2020) 1276.
[17] K. Shekhawat, V. K. Sharma, An extension of j-shaped distri-
bution with application to tissue damage proportions in blood,
Sankhya B 83 (2021) 548–574.

22