key: cord-0061883-48337vvq
authors: Almetwally, Ehab M.
title: The Odd Weibull Inverse Topp–Leone Distribution with Applications to COVID-19 Data
date: 2021-04-12
journal: Ann
DOI: 10.1007/s40745-021-00329-w
sha: 04761b2df3bde3424193a761da660b701c76ca94
doc_id: 61883
cord_uid: 48337vvq

This paper aims at defining an optimal statistical model for the COVID-19 distribution in the United Kingdom, and Canada. A combining the inverted Topp–Leone distribution and the odd Weibull family introduces a new lifetime distribution with a three-parameter to formulate the odd Weibull inverted Topp–Leone (OWITL) distribution. As a simple linear representation, hazard rate function, and moment function, this new distribution has several nice properties. To estimate the unknown parameters of OWITL distribution, maximum likelihood, least-square, weighted least-squares, maximum product spacing, Cramér–von Mises estimators, and Anderson–Darling estimation methods are used. To evaluate the use of estimation techniques, a numerical outcome of the Monte Carlo simulation is obtained.

Over the years, statistical lifetime distributions have gained a lot of coverage. Its interest has therefore evolved over time. Distribution theory researchers do this either by adding a new parameter to make the distribution of interest more versatile or even creating a new distribution family or modeling data in a variety of fields, including economics, engineering, reliability, and medical sciences Anake et al. [1] . Because of their applicability in many fields such as biological sciences, life test issues, medical, etc., inverted (or inverse) distributions are of great importance. The density and hazard ratio of the inverted distributions illustrate a distinct structure from the non-inverted distributions of conformation. The applications of inverted distributions have been discussed with many researchers, and the reader can refer to Abd AL-Fattah et al. [2] , Barco et al. [3] , Hassan and Abd-Allah [4] , Hassan and Mohamed [5] , Muhammed [6] , Chesneau et al. [7] , Usman and ul Haq [8] , Eferhonore et al. [9] among others. The modeling of COVID-19 data have been discussed with many researchers, and the reader can refer to Kumar [10] , Khakharia et al. [11] , Li et al. [12] , Liu et al. [13] , Wang [14] , Lalmuanawma et al. [15] and Bullock et al. [16] .

Hassan et al. [17] suggested the cumulative distribution function (CDF) and probability density function (PDF) of the inverted Topp-Leone distribution (ITL) distribution with form parameter > 0 as follows:

and, Kumar and Dharmaja [18] presented the exponentiated Kies distribution and some of its properties for this distribution. Dey et al. [19] . derived the product moments of the modified Kies distribution under Type II progressive censored sample, as well as an approximation of the distribution parameters. Bourguignon et al. [20] introduced the unusual Weibull-G family. Al-Babtain et al. [21] submitted a new distribution family based on the modified Kies (MK) distribution and the T-X family. A special case of the odd Weibull-G (OW) family with one parameter is the MK family. Almetwally et al. [22] introduced modified Kies inverted Topp-Leone distribution. If G(x; ) is the baseline CDF depending on a parameter vector , then the CDF of the OW family is defined by where is parameter vector ( , , ) . The corresponding PDF of (3) is given by

The motivation of the new distribution is modeling the COVID-19. We used the COVID-19 of the United Kingdom and Canada as real data to evaluate the use of the model techniques. The new cases or new deaths of COVID-19 data are discrete data (count). We used the daily mortality rate of COVID-19 for the United Kingdom and Canada. The daily mortality rate is continuous data.

The modulated three-parameter odd Weibull inverted Topp-Leone (OWITL) distribution, which has many desirable properties, which is obtained in this paper. The OWITL distribution has a very flexible PDF, can be positively skewed, symmetrical, and negatively skewed, and can allow tails to be more flexible. It is capable of modeling down, up, bathtub, upside-down bathtub, and reverse-J hazard rates monotonically. Also, it has a closed-form CDF and is very simple to manage, making the distribution a candidate for use in various fields, such as life testing, reliability, biomedical research,

and study of survival. Three real data applications show that certain conventional distributions with scale and shape parameters such as ITL, the Marshall-Olkin exponential, modified Kies exponential, inverse Weibull, inverse exponential, and inverse Rayleigh distributions are very competitive with the suggested distribution.

In alternative estimation methods, the maximum product spacing approach is used to estimate the continuous univariate model parameters as an alternative to the Maximum Likelihood method developed for complete sample by Cheng and Amin [23] and this developed to use under censored sample by Singh et al. [24] , Basu et al. [25] , Almetwally et al. [26] , El-Sherpieny et al. [27] , Alshenawy et al. [28, 29] . The leastsquare and weighted least-square methods are used to estimate the parameters of the beta distribution by Swain et al. [30] . Based on the discrepancy between CDF estimates and the empirical distribution function, the Cramér-von-Mises has been introduced by Cramér [31] and von Mises [32] . Luceño [33] , used the Cramér-von-Mises estimators to Fit the generalized Pareto distribution.

We plan to make a new extension bivariate OWITL based on copula in future studies, such as done in Almetwally et al. [34] , Muhammed and Almetwally [35] and Kim et al. [36] . We plan to discuss a new application for the OWITL distribution quest based on a censored sample such as done in Almetwally et al. [37] and Aslam et al. [38] .

The remainder of this paper is structured as follows: We get the OWITL distribution in Sect. 2. We address some of the mathematical properties of the OWITL distribution in Sect. 3. In Sect. 4, we get the OWITL distribution by an estimation process. In Sect. 5, OWITL distribution simulation results are obtained. Three implementations of real data analytics were obtained in Sect. 6. In Sect. 7, the paper is summarized and concluded.

Consider the ITL distribution of the positive scale factor and the CDF of Eqs. (1, 2) given (for x > 0 ). We define the OWITL distribution's CDF by inserting the ITL distribution's CDF into (3), such as:

where is parameter vector ( , , ) . The OWITL distribution's hazard rate (HR) feature is shown as 

that it is possible to increase, decrease and shape a bathtub for the HR feature in the OWITL distribution. One benefit of the distribution of OWITL over an ITL distribution is that the last of them can not model a phenomenon that shows increasing, decreasing shapes, failure rates of the bathtub, and therefore becomes more flexible to analyze data about lifetime.

For the OW family, we have a helpful linear representation and use it to provide a useful linear representation for the distribution of OWITL. A combination depiction of the OW family can be given as follows,

Using the ITL distribution's PDF and CDF, the last OWITL distribution equation can be rewritten as 

denotes the ITL density with parameter (h + 1).

A combination depiction of the CDF of OW family can be given as follows, By using Eq. 10, the CDF of OWITL distribution equation can be rewritten as

The OWITL distribution's quantile function, i.e.

, is derived as follows by inverting (5):

In particular, the first quartile Q1, the second quartile Q2, and the third quartile Q3 are obtained by setting Q = 0.25, 0.5, 0.75 , respectively, in Eq. (12).

According Hassan et al. [17] , the r th moment of X follows simply from Eq. (9) as

The r th incomplete moment of X can be obtained from (9) as

is the incomplete beta function.

This section uses six different estimation methods called: maximum likelihood, least-square, the maximum product of spacing, weighted least-square, Cramér-von Mises, and Anderson-Darling, to analyze the estimation problem of the OWITL distribution parameters.

Let x 1 , … , x n be a random sample with the parameters , and from an OWITL distribution. The log-likelihood function for the distribution of OWITL is given by

The partial derivatives of l( ) with respect to the model parameters , and are and (14)

It is possible to obtain the maximum likelihood estimation (MLE) of , , and by maximizing the last equation for , , and . By using the Newton-Rapshon method, the R packages can be used to optimize the log-likelihood function for obtaining the MLE.

To estimate the parameters of various distributions, the least-squares (LS) and weighted least-square (WLS) methods are used. Let x (1) < ⋯ < x (n) be a random sample with the , and parameters from the OWITL distribution. LS estimators (LSE) and WLS estimators (WLSE) of the , and distribution parameters of OWITL can be obtained by minimizing the following:

for WLSE with respect to , and . Furthermore, by resolving the nonlinear equations, the LSE and WLSE follow:

is a random sample of the size n, you can describe the uniform spacing of the OWITL distribution as: 

.

The Cramér-von-Mises (CVM) can be obtained for OWITL by minimizing the following function with respect to , , and , the CVM estimators (CVME) of the OWITL parameters , , and are obtained.

In addition, by resolving the nonlinear equations, the CVME as follows and (27)

In Anderson-Darling (AD), other forms of minimum distance estimators are the AD estimators (ADE). The ADE of the parameters of the OWITL is acquired by minimizing

Regarding , and , respectively. It is also possible to obtain the ADE by resolving the nonlinear equations.

In this portion, a simulation analysis evaluates the output of six different estimators of the OWITL parameters. For the different values of parameter = (0.5, 3) , = (0.5, 3) , and = (0.5, 3) , we consider the different sample sizes n = 50, 100, 150 . We create 10000 iteration of random samples for OWLTL distribution. We get the average values of relative bias (RB) and their corresponding mean square error (MSE) for each calculation.

The output of the different estimators is evaluated in terms of RB and MSE, i.e. those whose MSE values are closer to zero would be the most effective method of estimating. Simulation results are obtained via the R program. Tables 1 and 2 show the RB and MSE values for MLE, LS, WLS, MPS, CVM, and AD. Also, as the sample size increases in all situations, the mean depending on all estimation methods tends towards the true parameter values, suggesting that all estimators are asymptotically unbiased. If < 1 , and > 1 , the MPS is the best estimation methods in most times. If > 1 , and > 1 , the AD method is the best estimation methods in most times. If < 1 , < 1 and < 1 , the LS is the best estimation methods in most times. If < 1 , < 1 and < 1 , the LS is the best estimation methods in most times. If < 1 , > 1 and < 1 , the LS is the best estimation methods in most times. If , and increases and < 1 , the LS is the best estimation methods in most times (Figs. 3, 4 ).

This section is dedicated to demonstrate the potential of two real data sets for the OWITL distribution. Compared with other competitive models, OWITL delivery, namely: extended odd Weibull inverse Rayleigh (EOWIR) which is introduced by Almetwally [39] , generalized inverse Weibull (GIW) distribution which is introduced

by De Gusmao et al. [40] , exponential Lomax (ELo) distribution which is introduced by El-Bassiouny et al. [41] , modified Kies exponential (MKEx) which is introduced by Al-Babtain et al. [21] , and power Lomax (PL) distribution by Rady et al. [42] . For both models fitted on the basis of two real data sets, Tables 3 and 4 include Cramér-von Mises (W*), Anderson-Darling (A*) and Kolmogorov-Smirnov (KS) statistic values along with its P value. Furthermore, these tables include the parameters MLE and Standard Errors (SE) for the models considered (Fig. 5) .

The OWITL distribution has the highest P value and the lowest distance of the Kolmogorov-Smirnov(KS), W* and A* values in Tables 3 and 4 when compared to all other models used here to suit the COVID-19 results. Figures 6 and 7 show the empirical, histogram, QQ-plot, and PP-plot fit for the OWITL distribution of Canadian and United Kingdom COVID-19 results. These applications show that the OWITL model can yield better fit than some other distribution.

In order to classify the possible shapes behind these data of the unknown hrf, we plot the total time on test (TTT) plot in Fig. 5 (see Aarset [43] for further details on the use of TTT plots in data analysis). In Fig. 5 , since the blue line is convex, then concave, the unknown hrf probably presents a bathtub shape. Therefore, the OWITL distribution is appropriate to fit the data, where probTV is the accumulated probability distribution for time value and sorted_time_value is the vector time Value sorted of data. If data include outliers, we can use the Robust methods as least trimmed square, least median square, M, S, and MM, see Almongy and Almetwally [44, 45] . Artificial intelligence techniques can also be used see Olson et al. [46] , Shi et al. [47] and Tien [48] . Table 1 Table 2  Table 1 Mean 

We suggest a new three-parameter model in this paper, called the odd Weibull inverted Topp-Leone distribution, which can be denoted as an OWITL distribution. The distribution of OWITL is motivated by the wide use in life testing of the ITL model and provides more flexibility for evaluate lifetime data. Survival function, hazard function, linear, quantile representation, and OWITL distribution moments are given. We compare the methods of MLE, LSE, MPSE, WLSE, CVME, and ADE and conclude that the alternative methods of MLE are better than the MLE method. In the sense of statistics, we have two implementations of the OWITL distribution for COVID-19 data. The OWITL distribution estimation parameters are derived from MLE, LSE, MPSE, WLSE, CVME, and ADE. Estimation methods are used to estimate the parameters of the model and results of the simulation are given to test the performance of the model. The proposed model of two real-life data offers a consistently better fit than the distributions EOWIR, GIW, WLo, MKEx, and PL.

On a fractional beta-distribution

Inverted Kumaraswamy distribution: properties and estimation

The inverse power Lindley distribution

On the inverse power Lomax distribution

Parameter estimation for inverted exponentiated Lomax distribution with right censored data

On the inverted Topp Leone distribution

The inverted modified Lindley distribution

The Marshall-Olkin extended inverted Kumaraswamy distribution: theory and applications

Theoretical analysis of the Weibull alpha power inverted exponential distribution: properties and applications

Monitoring novel corona virus (COVID-19) infections in India by cluster analysis

Outbreak prediction of COVID-19 for dense and populated countries using machine learning

Culture versus policy: more global collaboration to effectively combat COVID-19

What are the underlying transmission patterns of COVID-19 outbreak? An age-specific social contact characterization

A call for caution in extrapolating chest CT sensitivity for COVID-19 derived from hospital data to patients among general population

Applications of machine learning and artificial intelligence for Covid-19 (SARS-CoV-2) pandemic: a review

Mapping the landscape of artificial intelligence applications against COVID-19

Statistical properties and estimation of inverted Topp-Leone distribution

The exponentiated reduced Kies distribution: properties and applications

Moments and estimation of reduced Kies distribution based on progressive type-II right censored order statistics

The Weibull-G family of probability distributions

A new modified Kies family: properties, estimation under complete and type-II censored samples, and engineering applications

A new inverted Topp-Leone distribution: applications to the COVID-19 mortality rate in two different countries

Estimating parameters in continuous univariate distributions with a shifted origin

Maximum product spacings method for the estimation of parameters of generalized inverted exponential distribution under Progressive Type II Censoring

Estimation of inverse Lindley distribution using product of spacings function for hybrid censored data

Adaptive type-II progressive censoring schemes based on maximum product spacing with application of generalized Rayleigh distribution

Progressive type-II hybrid censored schemes based on maximum product spacing with application to Power Lomax distribution

Product spacing of stress-strength under progressive hybrid censored for exponentiated-Gumbel distribution

Progressive type-II censoring schemes of extended odd Weibull exponential distribution with applications in medicine and engineering

Least-squares estimation of distribution functions in Johnson's translation system

On the composition of elementary errors: first paper: mathematical deductions

Wahrscheinlichkeit Statistik und Wahrheit

Fitting the generalized Pareto distribution to data using maximum goodness of fit estimators

Bivariate Weibull distribution: properties and different methods of estimation

Bayesian and non-Bayesian estimation for the bivariate inverse weibull distribution under progressive type-II censoring

Copula approach for developing a biomarker panel for prediction of dengue hemorrhagic fever

Maximum product spacing estimation of Weibull distribution under adaptive type-II progressive censoring schemes

Bayesian estimation of transmuted pareto distribution for complete and censored data

Extended odd Weibull inverse Rayleigh distribution with application on carbon fibres

The generalized inverse Weibull distribution

Exponential Lomax distribution

The power Lomax distribution with an application to bladder cancer data

How to identify a bathtub hazard rate

Robust estimation methods of generalized exponential distribution with outliers

Comparison between M estimation, S estimation, and MM estimation methods of robust estimation with application and simulation

Introduction to business data mining

Optimization based data mining: theory and applications

Internet of things, real-time decision making, and artificial intelligence

Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations

The authors are very grateful to the editor's board and reviewers for their care of the paper. The reviews are helpful to finalize the manuscript.

Funding The author received no specific funding for this study.

The data is included in Section 6. Application of Real Data Analysis.Code availability Function "mle2" of "bbmle" package in the R program has been used.

The author declares that they have no conflicts of interest to report regarding the present study. Authors' contributions Modeling for the COVID-19 distribution in the United Kingdom and Canada is studied. A new lifetime distribution with a three-parameter of odd Weibull inverted Topp-Leone is introduced. Important properties are studied. Parameter estimation was obtained by using different estimation methods.Ethical statements All of the followed procedures were in accordance with the ethical and scientific standards. This article does not contain any studies with human participants performed by the author.