The Kumaraswamy Weibull Geometric Distribution with Applications

In this work, we study the kumaraswamy weibull geometric (Kw-WG) distribution which includes as special cases, several models such as the kumaraswamy weibull distribution, kumaraswamy exponential distribution, weibull geometric distribution, exponential geometric distribution, to name a few. This distribution was monotone and non-monotone hazard rate functions, which are useful in lifetime data analysis and reliability. We derive some basic properties of the Kw-WG distribution including non-central rth-moments, skewness, kurtosis, generating functions, mean deviations, mean residual life, entropy, order statistics and certain characterizations of our distribution. The method of maximum likelihood is used for estimating the model parameters and a simulation study to investigate the behavior of this estimation is presented. Finally, an application of the new distribution and its comparison with recent flexible generalization of weibull distribution is illustrated via two real data sets.


Introduction
In analyzing lifetime data one often uses the Exponential, Rayleigh, Linear Failure Rate or Weibull distributions. These distributions have several popular properties and nice physical interpretations which make them quite useful. The Weibull distribution (Weibull, 1951) has been used in many different fields of applications. The hazard rate function (hf) of the Weibull distribution can only be increasing, decreasing or constant. So it cannot be used to model lifetime data with a bathtub or other shape of hazard function, such as human mortality and machine life cycles. For several years, researchers have been developing various extensions of the Weibull distribution, with number of parameters ranging from 2 to 5. For example Exponentiated Weibull (EW) Distribution (Manal and Fathy, 2003), Beta-Weibull (BW) distribution (Lee et al., 2007), Kumaraswamy Weibull  where G(x) is a cdf with its corresponding pdf, g(x). Clearly, for a=b=1, we obtain the main distribution. The parameters a and b control the skewness and tail weights. The form of this class of distributions is simpler than the Beta-G class (Eugene et al., 2002) because it does not involve incomplete beta function. Nadarajah et al. (2012) obtained general results about this class of distributions.
In this paper, a more flexible five parameter generalization of Weibull distribution based on Kumaraswamy-G class and Weibull Geometric distribution, called Kumaraswamy Weibull Geometric (Kw−WG), is introduced. A comprehensive description of some of its mathematical properties is presented. The pdf of this model includes decreasing, right and left skew uni-bimodal shape and the hf of this distribution contain increasing, decreasing, unimodal, bathtube and non-monotone shape. The Kw-WG distribution includes some well-known class of distribution as special cases such as Kw-W, Kw-Rayleigh (Gomez et  al,  This article is organized as follows. In Section 2, the new distribution with its pdf is proposed. Further, the distributional properties of the new distribution, such as the cdf, survival and hazard rate functions, moment generating function, non-central and descending factorial moments, mean deviation, Bonferroni and Lorenz curves, mean residual life and mean inactivity time, Renyi and q entropy functions and order statistics with its moments are discussed. Section 3 deals with the characterizations of this model based on truncated moments, hazard function, reversed hazard function and certain function of the random variable. In Section 4, the maximum likelihood estimation of the parameters and the Fisher information matrix are discussed and a simulation study to investigate behavior of the estimators is done. Applications of the proposed model are illustrated in Section 5. Concluding remarks are given in Section 6.

The Kumaraswamy-Weibull Geometric distribution
The cdf of is defined by (1) where and are parameters. Its pdf has the form (2) Note that, is a scale parameter and the other positive parameters a,b,p and c are shape parameters. The graphs of pdf in (2), for selected parameters values are given in Figure 1. This distribution is more flexible than Weibull Geometric distribution and can be model decreasing, right and left skew unimodal and bimodal data sets. If X is a random variable with pdf (2), we write X~ . The survival and hazard rate functions corresponding to (2) are  Therefore, X has the Kw − WG distribution given by (1). The distributions which are sub-models of the Kw − WG distribution are listed in Table  1.
Also, the asymptotics of cdf, pdf and hf as x → ∞ are given by and These equations show the effect of parameters on tails of distribution.

Extreme Value
If is a random sample from Kw−WG distribution and if denotes the sample mean, then by usuall central limit theorem approaches the standard normal distribution as n → ∞. One may be interested in the asymptotic of the extreme values and For Kw − WG distribution, it can be seen that can be rewritten as the following series representation

Moments and generating function
Some of the most important characteristics of a distribution can be studied through moments. For a random variable X having density function (2), is obtained by using Equation (4) and Gamma integral as (6) The skewness and kurtosis measures can be calculated from the ordinary moments using well−known relationships. Graphical representation of these quantities for some choices of parameter b as function of a, by fixing and , are given in Figures 3.
The central moments ( ) and cumulants ( ) of X can be obtained from Equation (6)   ( ) where is the Stirling number of the first kind. Thus the factorial moments of X are given by The moment genrating function (mgf) of X, can be expressed as Thus the mgf of is

Quantile function
The quantile function, say , of the Kw WG distribution obtaines by inverting Equation (1) as (7) We can simulate data from the distribution by where u has the uniform U(0,1) distribution.

Mean deviations
The amount of scatter in X is evidently measured to some extent by the totality of deviations from the mean and median. If X has the Kw− WG distribution (1), we can derive the mean deviations about the mean and about the median M as and respectively. The median M is obtained from Equation (7)  where is easily obtained from Equation (1) and From Equation (4) and using the incomplete gamma function we can obtain (9) Equation (9) is the basic value to compute the mean deviations and in Equation (8).
It can also be used to determine Bonferroni and Lorenz curves. These curves have applications not only in economics in the study of income and poverty, but also in other fields like reliability, demography, insurance and medicine.

Mean residual life and mean inactivity time
The mean residual life has many applications in biomedical sciences, life insurance, maintenance and product quality control, economics and social studies, demography and product technology (see Lai and Xie, 2006). The MRL is given by for t > 0, and it represents the expected additional life length for a unit, which is alive at age t.
The MRL of X can be computed as where J(t) is Equation (9).
The mean inactivity time defined by (for t > 0) represents the waiting time elapsed since the failure of an item on condition that this failure had occurred in (0,t). The MIT of X is given by where J(t) is Equation (9).

Entropy
The entropy of a random variable X is a measure of the uncertainty variation. The Renyi entropy is defined by where and . Similarily, is expanded as follow Therefore, the Renyi entropy for the Kw-WG distribution is given by (11) The q-entropy is defined by Where ( and ), follows from (11) as

Order statistics and moments
In this section, we derive closed-form expressions for the pdf of the rth order statistic of X. Let be a random sample from the Kw − WG distribution with cdf and pdf given by (1) and (2), respectively. Let denote the order statistics obtained from this sample. The pdf of , say , is given by where F(x) and f(x) are the cdf and pdf of X given by (1) and (2), respectively, and B(.,.) is the beta function. Similarily, expands as where The rth moment of the ith order statistic can be obtained from the following result:

Characterizations
Characterizations of distributions is an important research area which has recently attracted the attention of many researchers. This section deals with various characterizations of Kw-WG family of distributions. These characterizations are based on: (i) a simple relationship between two truncated moments; (ii) the hazard function; (iii) the reverse hazard function; (iv) certain functions of the random variable. It should be mentioned that for characterization (i) the cdf need not have a closed form. We present our characterizations (i) − (iv) in four subsections.

Characterizations based on two truncated moments
In this subsection we present characterizations of Kw-WG distribution in terms of a simple rela-tionship between two truncated moments. This characterization result employs a theorem due to Gl¨anzel (1987) see Theorem 1 below. Note that the result holds also when the interval H is not closed. Moreover, as mentioned above, it could be also applied when the cdf F does not have a closed form. As shown in [10], this characterization is stable in the sense of weak convergence.

Characterization based on hazard function
It is known that the hazard function, , of a twice differentiable distribution function, F, satisfies the first order differential equation (12) For many univariate continuous distributions, this is the only characterization available in terms of the hazard function. The following characterization establish a non-trivial characterization of Kw-WG distribution, for a = 1, in terms of the hazard function, which is not of the trivial form given in (12).

Proposition 2.
Let : Ω → (0, ∞) be a continuous random variable. For a = 1,the pdf of X is (2) if and only if its hazard function satisfies the differential equation (13) with the initial condition for c > 1.
Proof. If X has pdf (2), then clearly (13) holds. Now, if (13) holds, then or which is the hazard function of the Kw-WG distribution for a = 1.

Characterization in terms of the reverse (or reversed) hazard function
The reverse hazard function, of a twice differentiable distribution function, F, is defined as Proposition 3. Let : Ω → (0, ∞) be a continuous random variable. For b = 1, the pdf of X is (2) if and only if its reverse hazard function satisfies the differential equation (14) Proof. If X has pdf (2), then clearly (14) holds. Now, if (14) holds, then or which is the reverse hazard function of the Kw-WG distribution. The following propositions have already appeared in (Hamedani, 2013) , so we will just state them here which can be used to characterize Kw-WG distribution. there are other suitable functions ψ, we choose the above ones for simplicity.

Maximum likelihood estimation
We calculate the maximum-likelihood estimates (MLEs) of the parameters of the Kw-WG dis-tribution from complete samples only. Let Let be a random sample of size n from the Kw − WG(a,b,p,c,λ) distribution. The log-likelihood function for the vector of parameters can be written as (15) The log-likelihood can be maximized either directly by using the R program or by solving the nonlinear likelihood equations obtained by differentiating Equation (15). In this section, we investigate the behavior of the ML estimators for a finite sample size (n). Simulation study based on Kw−WG(a,b,p,c,λ) distribution is carried out. The random variables are generated by using quantile technique presented in section 2.5 from Kw − WG(a,b,p,c,λ). A simulation study consisting of following steps is being carried out for each (a,b,p,c,λ) such as (0.5,0.   Table 2 indicate that the estimates are stable and are more close to the true values when the sample sizes increased.  In this section, we present an application of the Kw − WG distribution using two real data sets. The first data set is given by Raqab and Kundu (2009) on the gauge lengths of 20 mm and consists of n = 74 observations. Also, these data set is used by Nofal et al. (2015) and Afify et al. (2016). We use the same data to compare the Kw − WG model with some rival models. The second data set is an uncensored data set from Murthy et al. (2004) (page 180) consisting of 50 observation on failure times of 50 Components (per 1000 hours). These data were also analyzed by Merovci and Elbatal (2013).
In the applications, the information about the hazard shape can help in selecting a particular model. For this aim, a device called the total time on test (TTT) plot (Aarset, 1987) is useful. The TTT plot is obtained by plotting where r = 1,...,n and (i = 1,...,n) are the order statistics of the sample, against r/n. If the shape is a straight diagonal the hazard is constant. It is convex shape for decreasing hazards and concave shape for increasing hazards. The bathtub-shaped hazard is obtained when the first convex and then concave and for bimodal shape hazard, the TTT plot is first concave and then convex. The TTT plot for both datasets presented in Figure 4.
These figures indicates that first and second dataset has increasing hazard and decreasing failure rate function.
Therefore, the proposed Kw-WG model can be used to fit these data, since it can be model data with increasing and decreasing shapes of failure rate functions. The MLE of parameters, maximized log-likelihood function, Cramer-von Mises ( ) and AndersonDarling ( ) statistics are determined for fitting distributions.  Table 3 and Table 4 include the MLEs of the model parameters, their corresponding standard errors (SEs) and the values of maximized log-likelihood function, and . Figures 5 and 6 displays histogram and estimated cdf of dataset with fitted Kw-WG distribution.   Table 4: MLEs, their SEs (in parentheses), maximized log-likelihood, , and for the fitted models to the second data set  It is seen that the proposed Kw−WG model provides the best fit for both data sets when considering maximized log likelihood, Anderson-Darling and Cramer-von Mises goodness of fit statistics.

Conclusion
This paper proposed a new five parameter lifetime distribution by using Kumaraswamy-G class of distributions and Weibull Geometric distribution. It contains a number of known special submod-els such as Kumaraswamy Weibull distribution, Kumaraswamy Exponential distribution, Weibull Geometric distribution, Exponential Geometric distribution and etc. We have studied many statistical properties of Kw − WG distribution including its probability density function, explicit algebraic expressions of survival and hazard functions, mean deviation, order statistics and its mo-ments, entropy, moment generating function and its characterizations. The maximum likelihood estimation is discussed and a simulation study for its behavior is done. It is suitable for increasing, decreasing, bathtub shaped, and non monotone shaped hazard rates, indicating its flexibility for modeling lifetime data with various shaped hazard rate functions. Accordingly, we expect that the new distribution may attract wider applications in reliability, biology, and lifetime data analysis.