Bayesian Prediction Intervals Based on Type-I Hybrid Censored Data from the Lomax Distribution under Step-Stress Model

Rabie, Abdalla; Ahmad, Abd EL-Baset A.; Fawzy, Mohamad A.; Aloafi, Tahani A.

doi:https://doi.org/10.1155/2022/2801582

Journal of Mathematics

On this page

Abstract Introduction Conclusion Notation Data Availability Conflicts of Interest Acknowledgments References Copyright Related Articles

Research Article | Open Access

Volume 2022 | Article ID 2801582 | https://doi.org/10.1155/2022/2801582

Bayesian Prediction Intervals Based on Type-I Hybrid Censored Data from the Lomax Distribution under Step-Stress Model

Abdalla Rabie,¹Abd EL-Baset A. Ahmad,²Mohamad A. Fawzy,^3,4and Tahani A. Aloafi⁵

Academic Editor: Ali Sajid

Received23 Oct 2022

Revised06 Dec 2022

Accepted07 Dec 2022

Published23 Dec 2022

Abstract

The Bayesian prediction of future failures from Lomax distribution is the subject of this research. The observed data is censored using a Type-I hybrid censoring scheme under a step-stress partially accelerated life test. There are two types of sampling schemes considered: one-sample and two-sample. We create predictive intervals for failure observations in the future. Bayesian prediction intervals are constructed using MCMC algorithms. After all, two numerical examples, simulation study and a real-life example are provided for both one-sample and two-sample methods for the purpose of illustration.

1. Introduction

In highly industrial products, it is difficult to obtain sufficient information about the product to test its validity, so it is appropriate to use an acceleration life test to obtain enough information with saving time and high cost. There are two types of accelerated life tests: the fully accelerated life test, in which it is assumed that the acceleration factor is known and there must be a mathematical model that determines the relationship between the life span and stress. Another one is the partially accelerated life test which is applied when the relationship between the life span and stress is unknown and is applied throughout step stress and constant stress. Many authors have written on this topic, see for example, Saeed, and Hanieh [1]; Lone et al. [2] and Mohammad and Mohammad [3] and Nagy et al. [4].

It is important to use past data to predict future data in many experiments of life tests for units or products to improve product efficiency. One of the fascinating subjects in real-world reliability issues is prediction. Recently, there has been a lot of discussion about various prediction techniques.

In multiple fields as healthcare, finance, technology, and education, Bayesian prediction of future data plays a critical role. The Bayesian approach, according to Geisser [5]; can be used to tackle the problem of prediction. See also AL-Hussaini et al. [6, 7], Ahmadi et al. [8], Ahmad [9], Ahmad et al. [10], Ateya [11], Balakrishnan and Shafay [12], Kundu and Howlader [13], and Singh et al. [14], Ahmad et al. [15] and Corcuera and Giummole [16].

The most common type of prediction is predictive intervals. They are not the same as the commonly used confidence intervals and tolerance regions, which are intended to deal with unknown population parameters. A predictive interval is a range of values that uses the results of a previous sample to predict the outcomes of a future sample with a given probability.

Prediction concerns can be divided into two categories:(1)One-sample prediction Let and represent the known sample and a future sample, respectively. A one-sample prediction scheme entails predicting the future sample for .(2)Two-sample prediction Let and denote the informative sample of size and a future sample of size , respectively. The two samples are believed to be unrelated and drawn from the same distribution. The prediction of the future sample is part of a two-sample prediction issue.

For more details about one and two sample methods, one can refer to Prakash [17]; Wu and Gui [18]; and Mohie El-Din et al. [19].

1.1. The Model Description and Censoring Scheme

The Lomax distribution was introduced by Lomax [20] to analyse business failure data. Several authors have used Lomax distribution, for example, Howlader and Hossain [21], Abd-Ellah [22], Abd-Elfattah et al. [23], Hassan and Al-Ghamdi [24], among others. The cumulative distribution function (CDF) and the probability density function (PDF) of Lomax distribution are written, respectively, in the formswhere, and are the shape and scale parameters, respectively.

By mixing Type-I and Type-II censoring schemes, we obtain hybrid censoring scheme (HCS) in which the test is ended after obtaining a fixed number of failures from items or after reaching a pre-determined time T. So, we can observe the following types of HCSs.(i)Type-I HCS: Epstein [25] considered a HCS in which the life-testing experiment is terminated at a random time , where and are known before.(ii)Type-II HCS: In this type the experiment is finished at a random time , this ensures at least of failures are obtained.

In this paper, we consider Type-I HCS which includes the following types of censored data: Case 1: , if , and pre-specified number of failures occur before the censoring time . Case 2: , if , and only number of failures occur before the pre-specified censoring time .

The rest of this paper is organized as follows: Section 2 gives the test method and procedure. In Section 3, the likelihood function and the posterior PDF are stated. Section 4 is devoted for Bayesian one-sample prediction scheme and MCMC method. Bayesian two-sample prediction and MCMC technique are presented in Section 5. Section 6 presents numerical computations involving numerical examples and simulation study, as well as real-life example. The findings are displayed in four tables then concluding the results and methodology in Section 7.

2. The Test Method

In this section, we explain the fundamental assumptions and test procedures for the step-stress partially accelerated life test (SSPALT) model based on the Type-I hybrid censoring method.

2.1. Fundamental Presumptions

(1)There are two stress levels: and (design and high).(2)The life distribution of the test unit is Lomax distribution for any level of stress.(3)An item’s entire lifetime is calculated as follows: where, is an item lifetime at normal use condition and , is the accelerating factor. This model discussed by Degroot and Goel [26] and referred to as tampered random variable (TRV).(4)The lifetimes of the test items are independent and identically distributed random variables.

2.2. Test Procedure

(1)All test items are put through a normal environment.(2)If number failures does not fail by a predetermined time in normal use, it is switched to accelerated mode and run until it fails or the censorship time is reached.

Based on the above assumptions, the CDF and PDF of a total lifetime of test item take the following forms, respectively,where, and

In Type-I HCS, the life-testing experiment is stopped according to the rule , that is the test is terminated at a predetermined time if the failure obtained after , or as soon as the failure occurs, where . Thus,

We may obtain a random number of failures given by

3. The Likelihood Function and Posterior PDF

The likelihood function under Type-I HCS according to the previous assumptions can be written as:

For Case 1, in this case the prefixed number out of items is obtained at a normal use condition, so, this case will be discarded.

And for Cases 2 and 3, the likelihood function under Type-I HCS can be combined and written as:where, , and .

We consider the likelihood function in Cases 2 and 3 only, as it runs at accelerated use condition.

We suggest the prior PDF for the parameters and to be and , respectively; because the gamma prior is wealthy enough to cover the prior belief of the experimenter, so, many authors used it, see Pak and Mohammad [27]; Okasha [28]; Abd-Elfattah et al. [23]; which expressed as follows:

It should be pointed out that, the empirical Bayes method can be used to estimate the hyper parameters if they are unknown using previous samples, see, for example, Maritz and Lwin [29]. As an alternative, the hierarchical Bayes approach, which employs a suitable prior for the hyper parameters, could be utilized, see Bernardo and Smith [30].

The prior PDF for the acceleration factor is a non-informative prior given by:

The joint prior PDF of and , can be written as follows:

The majority of published papers consider that the choice of prior distribution is the researcher’s belief and the assumption of independence between parameters is the best in terms of ease of calculations and good results, and there is no way to know independence or dependence between them. Also, they are specified as independent when you do not want to assume that they are prior informative about each other. That is, knowing the value of one would not change your mind about any of the others, before seeing any data, this assumption was chosen when we determined the prior distribution and this is more understandable. In addition, choosing another prior distribution or dependent parameters will increase the complexity and difficulty of mathematical equations. Furthermore, the independent gamma priors are relatively simple and concise, which may not yield much complex inferential and computational issues. Also, in many practical situations, although the dependent prior models seem more attractive, yet the dependent property between parameters cannot be justified from a statistical perspective due to historical information and expert experience where such prior data/information may be very rare. Therefore, independent priors are more popular in statistics under the Bayesian procedure for the sake of simplicity, see EL-Sagheer et al. [31]; and Nassar et al. [32].

The joint posterior PDF of and , given , can be written from (7) and (11), aswhere, a normalizing constant is

The conditional posterior PDF of and , are, respectively, given by

4. Bayesian One-Sample Prediction

Let the first order statistics, , , have been observed and we want to predict the future order statistics, , . The conditional PDF of the future order statistics given informative order statistics , can be written as:

substituting (3) and (4) in (17), we getwhere, .

By multiplying (12) by (18) and then integrating with respect to , the predictive PDF of , given the informative order statistics , is given by

By substituting (12) and (18) in (19), we obtain

A , two-sided Bayesian predictive interval for , is given bywhere and are lower and upper bounds of Bayesian prediction for , are obtained by solving the following two equations numerically

4.1. Bayesian Prediction: One-Sample Scheme via MCMC

The predictive PDF (20), is approximated by using the MCMC method as follows:where, , , are generated from the posterior PDF (12).

A , two-sided Bayesian predictive interval, , of observations can be calculated from solving the following nonlinear equations for a given

5. Bayesian Two-Sample Prediction

Let be the order statistics from a future random sample of size and independent of the informative sample . The future sample and the informative sample are following the Lomax SSPALT model under Type-I HCS. We wish to predict the first future order statistics of the future sample .

The marginal PDF of the order statistics from is expressed as:where, .

The predictive PDF of , say is given as:where, is the joint posterior PDF of as given in (12). By substituting from (12) and (25) in (26), we get

A , two-sided Bayesian predictive interval for , is written as:where, bounds (L, U) of , are computed by numerical solution of the two equations for a given

5.1. Bayesian Two-Sample Prediction under MCMC

The predictive PDF (26), is approximated by applying the MCMC as follows:

A , two-sided Bayesian predictive interval, , of the future observations are obtained by solving the following nonlinear equations

6. Numerical Computations

Here, we present two examples to illustrate one-sample and two-sample techniques, as well as a simulation study and real-life data for the discussed methodology, is stated.

6.1. Numerical Examples

6.1.1. One-Sample Scheme

(1)For known values of of hyper parameters , generate and from equations (8), (9) and (10), respectively.(2)By using generated values of and , a random sample of size 30 is simulated from (4). The first informative ordered failures are obtained and listed as follows: 0.0113981, 0.0188695, 0.0638502, 0.06688, 0.0716792, 0.217725, 0.22276, 0.224412, 0.224614, 0.227522, 0.228146, 0.233232, 0.237106, 0.240437, 0.246286, 0.25205, 0.259451, 0.260386, 0.270926, 0.278369.(3)Using these informative data in (25) with 0.95 CI, the lower and upper bounds of the next predicted failure are (0.27903, 0.400507) and for the last predicted failure are (0.279084, 0.402948).

6.1.2. Two-Sample Scheme

(1)For given values of of hyper parameters , generate and from equations (8), (9) and (10), respectively.(2)By using generated values of and , a random sample of size 20 is simulated from (3). The first informative ordered failures are obtained and listed as follows: 0.00167013, 0.157792, 0.193577, 0.19366, 0.207637, 0.219493, 0.226598, 0.242315, 0.271873, 0.294316, 0.334488, 0.603327, 0.651955, 0.70614, 0.743701.(3)Using these informative data in (31) with 0.95 CI, the lower and upper bounds of the first four observations of a non-informative sample are (0.00129258, 0.191083), (0.00136062, 0.201295), (0.00143622, 0.21266) and (0.00152071, 0.225385), respectively.

6.2. Simulation Study

In this part of the paper, we explain the MCMC algorithm that used for computing Bayesian bounds for future samples in the case of one-sample and two-sample approaches. We determine values of and choose . Using values of prior parameters we generate initial values of the parameters , and from their prior PDF. Based on the generated , and , we generate a Lomax Type-I HC sample using SSPALT model with different sizes and censoring values based on the inverse function technique given bywhere, denotes a number obtained from randomly. From Metropolis algorithm, is generated from (14), from (15) and from (16) using normal distribution as a proposal distribution. Because the predictive PDF (21) cannot be produced in a closed form for one-sample prediction, it is approximated using the MCMC approach as described in (22). A 95% Bayesian predictive intervals for Y_s::n, s=r∗ + 1, r∗ + 2, …, n are obtained by solving the two nonlinear Eqs.(23). For two-sample prediction, the predictive PDF (26), which cannot be produced in a closed form, is estimated using the MCMC approach as described in (28). A 95% Bayesian predictive intervals for , are obtained by solving the two nonlinear equation (29).

All results are derived by Mathematica 8 programming language software. The results of one-sample scheme with different values of and are listed in Table 1. The results of two-sample case with the same values of and are obtained and displayed in Table 2. The numerical results are computed by MATHEMATICA 8 codes, such as: (FindRoot, NMaximize, NIntegrate and RandomReal).

6.3. Illustrative Example

In this section, we present a real life example, the dataset was initially considered by Chhikara and Folks [33]. It represents 46 repair times (in hours) for an airborne communication transceiver. The Lomax distribution has been fitted on this data set by Singh et al. [34]; and they stated that the Lomax distribution can be used for analyzing this data set. The data set is ordered and listed as follows:

0.2, 0.3, 0.5, 0.5, 0.5, 0.5, 0.6, 0.6, 0.7, 0.7, 0.7, 0.8, 0.8, 1.0, 1.0, 1.0, 1.0, 1.1, 1.3, 1.5, 1.5, 1.5, 1.5, 2.0, 2.0, 2.2, 2.5, 2.7, 3.0, 3.0, 3.3, 3.3, 4.0, 4.0, 4.5, 4.7, 5.0, 5.4, 5.4, 7.0, 7.5, 8.8, 9.0, 10.3, 22.0, 24.5.

By applying SSPALT on this data set, by considering this data represents lifetimes of units put on life testing, we can conclude the following:(1)At , , and , we have 13 items are failed at throughout normal usage. After changing mode to acceleration case, the test is terminated at , that is, 32 items are failed at .(2)At , , and , there have been 20 items are failed at throughout normal usage. After switching to acceleration case, the test is terminated at , that is, 35 items are failed at .

Based on SSPALT model with Type-I HCS, one-sample Bayesian predictive intervals for order statistic , are reported in Table 3 and from the same sample two-sample Bayesian predictive intervals for order statistic , , are reported in Table 4.

7. Conclusion

Based on SSPALT under Type-I HCS of order statistics data from Lomax distribution, Bayesian prediction bounds for future observations are obtained by using one-sample and two-sample prediction techniques. The MCMC method is used to get Bayesian predictive intervals. We can state that using SSPALT with Type-I HCS, which has the advantage of having pre-determined experiment time, maximizes this advantage and speed up obtaining more information. Furthermore, one can see that SSPALT improves the disadvantage of this scheme, which is represented in the lack of failures in the specified time for the experiment. Finally, some numerical examples and simulation study are given to illustrate the results, and we observe the following remarks:

From Tables 1 and 2 (one-sample and two-sample cases), we noted that the Bayesian predictive intervals were affected by changing the values of and as follows:(1)When increasing and with the same , the length of Bayesian predictive intervals decreases.(2)The length of Bayesian predictive intervals, for fixed values of and , tends to be shorter when and increase, except for a few cases. This may be due to fluctuations in data. From Tables 3 and 4 (one-sample and two-sample cases), we observed that:(3)When values of and get larger the length of Bayesian predictive intervals becomes smaller.

Notation

:	A specific maximum time of the experiment
:	A random termination time of experiment
:	A total lifetime of an item in a step-stress model
y _r:n:	A time in which item fail
:	A specified time at which the stress is changed from to
:	A number of units are failed before time at stress level
:	Units number that fail before time at stress level
:	A number of units that fail before time at the stress level where is.

Data Availability

The data are generated by simulation, and the real data was considered by Chhikara and Folks [30].

Conflicts of Interest

The authors declare that they have no conflicts of interest.

Acknowledgments

This research was funded by Taif University Researchers Supporting Project number (TURSP-2020/321), Taif University, Taif, Saudi Arabia.

References

A. Saeed and P. Hanieh, “Bayesian prediction for censored data from the kumaraswamy distribution based on constant-stress accelerated life test model and its application in ceramic materials,” Journal of Quality Engineering and Managment, vol. 11, no. 1, pp. 15–32, 2021.
View at: Google Scholar
S. A. Lone, H. Panahi, and I. Shah, “Bayesian prediction interval for a constant-stress partially accelerated life test model under censored data,” Journal of Taibah University for Science, vol. 15, no. 1, pp. 1178–1187, 2021.
View at: Publisher Site | Google Scholar
A. Mohammad and Z. R. Mohammad, “Prediction of censored weibull lifetimes in a simple step-stress plan with khamis-higgins model,” STATISTICS, OPTIMIZATION AND INFORMATION COMPUTING, vol. 10, no. 3, pp. 0–19, 2021.
View at: Google Scholar
M. Nagy, M. H. Abu‐Moussa, A. F Alrasheedi, and A. Rabie, “Expected Bayesian estimation for exponential model based on simple step stress with Type‐I hybrid censored data.,” Mathematical Biosciences and Engineering, vol. 19, no. 10, pp. 9773–9791, 2022.
View at: Google Scholar
S. Geisser, Predictive Inference: An Introduction, Champman and Hall, London, UK, 1993.
E. K. Al-Hussaini and A. A. Ahmad, “On Bayesian interval prediction of future records,” Test, vol. 12, no. 1, pp. 79–99, 2003.
View at: Publisher Site | Google Scholar
E. K. Al-Hussaini and A. A. Ahmad, “On Bayesian predictive distributions of generalized order statistics,” Metrika, vol. 57, no. 2, pp. 165–176, 2003.
View at: Publisher Site | Google Scholar
J. Ahmadi, S. M. T. K. MirMostafaee, and N. Balakrishnan, “Bayesian prediction of k-record values based on progressively censored data from exponential distribution,” Journal of Statistical Computation and Simulation, vol. 82, no. 1, pp. 51–62, 2012.
View at: Publisher Site | Google Scholar
A. A. Ahmad, “On Bayesian interval prediction of future generalized-order statistics using doubly censoring,” Statistics, vol. 45, no. 5, pp. 413–425, 2011.
View at: Publisher Site | Google Scholar
A. A. Ahmad, M. A. Fawzy, and H. Ouda, “Bayesian prediction of future observations from weighted exponential distribution constant-stress model based on type-II hybrid censored data,” Communications in Statistics - Theory and Methods, vol. 50, no. 12, pp. 2732–2746, 2021.
View at: Publisher Site | Google Scholar
S. F. Ateya, “Prediction under generalized exponential distribution using MCMC algorithm,” International Mathematical Forum, vol. 6, no. 63, pp. 3111–3119, 2011.
View at: Google Scholar
N. Balakrishnan and A. R. Shafay, “One- and two-sample Bayesian prediction intervals based on type-II hybrid censored data,” Communications in Statistics - Theory and Methods, vol. 41, no. 9, pp. 1511–1531, 2012.
View at: Publisher Site | Google Scholar
D. Kundu and H. A. Howlader, “Bayesian inference and prediction of the inverse Weibull distribution for Type-II censored data,” Computational Statistics & Data Analysis, vol. 54, no. 6, pp. 1547–1558, 2010.
View at: Publisher Site | Google Scholar
S. K. Singh, U. Singh, and V. K. Sharma, “Bayesian prediction of future observations from inverse Weibull distribution based on type-II hybrid censored sample,” International Journal of Advanced Statistics and Probability, vol. 1, no. 2, pp. 32–43, 2013.
View at: Publisher Site | Google Scholar
K. E. Ahmad, Z. F. Jaheen, and H. S. Mohammed, “Bayesian prediction based on type-I censored data from a mixture of Burr type XII distribution and its reciprocal,” Statistics, vol. 46, no. 6, pp. 787–797, 2012.
View at: Publisher Site | Google Scholar
J. M. Corcuera and F. Giummolè, “A generalized Bayes rule for prediction,” Scandinavian Journal of Statistics, vol. 26, no. 2, pp. 265–279, 1999.
View at: Publisher Site | Google Scholar
G. Prakash, “Bayes prediction on optimum SS-PALT in generalized inverted exponential distribution: a two-sample approach,” Austrian Journal of Statistics, vol. 51, pp. 1–15, 2022.
View at: Publisher Site | Google Scholar
M. Wu and W. Gui, “Estimation and prediction for Nadarajah-Haghighi distribution under progressive type-II censoring,” Symmetry, vol. 13, no. 6, p. 999, 2021.
View at: Publisher Site | Google Scholar
M. M. Mohie El-Din, A. Sadek, and M. Nagy, “Bayesian Estimation and two-sample prediction based on unified hybrid censored sample,” J. Stat. Appl. Pro., , vol. 5, no. 3, pp. 439–448, 2016.
View at: Publisher Site | Google Scholar
K. S. Lomax, “Business failures: another example of the analysis of failure data,” Journal of the American Statistical Association, vol. 49, no. 268, pp. 847–852, 1954.
View at: Publisher Site | Google Scholar
H. A. Howlader and A. M. Hossain, “Bayesian survival estimation of Pareto distribution of the second kind based on failure-censored data,” Computational Statistics & Data Analysis, vol. 38, no. 3, pp. 301–314, 2002.
View at: Publisher Site | Google Scholar
A. H. Abd-Ellah, “Bayesian one sample prediction bounds for the Lomax distribution,” Indian Journal of Pure and Applied Mathematics, vol. 3, pp. 101–109, 2003.
View at: Google Scholar
A. M. Abd-Elfattah, F. M. Alaboud, and A. H. Alharby, “On sample size estimation for Lomax ditribution,” Aust. J. Basic Appl. Sci., , vol. 1, pp. 373–378, 2007.
View at: Google Scholar
A. S. Hassan and A. S. Al-Ghamdi, “Optimum step stress accelerated life testing for Lomax distribution,” Journal of Applied Sciences Research, vol. 5, pp. 2153–2164, 2009.
View at: Google Scholar
B. Epstein, “Truncated life tests in the exponential case,” The Annals of Mathematical Statistics, vol. 25, no. 3, pp. 555–564, 1954.
View at: Publisher Site | Google Scholar
M. H. Degroot and P. K. Goel, “Bayesian estimation and optimal designs in partially accelerated life testing,” Naval Research Logistics Quarterly, vol. 26, no. 2, pp. 223–235, 1979.
View at: Publisher Site | Google Scholar
A. Pak and R. M. Mohammad, “Estimating the parameters of lomax distribution from imprecise information,” Journal of Statistical Theory and Applications, vol. 17, no. 1, pp. 122–135, 2018.
View at: Google Scholar
H. M. Okasha, “E-Bayesian estimation for the Lomax distribution based on type-II censored data,” Journal of the Egyptian Mathematical Society, vol. 22, no. 3, pp. 489–495, 2014.
View at: Publisher Site | Google Scholar
J. S. Maritz and T. Lwin, Empirical Bayes Methods, Chapman & Hall, London, UK, 1989.
J. M. Bernardo and A. F. M. Smith, Bayesian Theory, ATLANTISSPS, New York, NY, USA, 1994.
R. M. El-Sagheer, M. S. Eliwa, K. M. Alqahtani, and M. El-Morshedy, “Asymmetric randomly censored mortality distribution: bayesian framework and parametric bootstrap with application to COVID-19 data,” Journal of Mathematics, vol. 2022, Article ID 8300753, 14 pages, 2022.
View at: Publisher Site | Google Scholar
M. Nassar, S. Dey, L. Wang, and A. Elshahhat, “Estimation of Lindley constant-stress model via product of spacing with Type-II censored accelerated life data,” Communications in Statistics - Simulation and Computation, pp. 1–27, 2021.
View at: Publisher Site | Google Scholar
R. S. Chhikara and J. L. Folks, “The inverse Gaussian distribution as a lifetime model,” Technometrics, vol. 19, pp. 461–468, 1977.
View at: Google Scholar
S. K. Singh, U. Singh, and A. S. Yadav, “Bayesian estimation of Lomax distribution under type-II hybrid censored data using Lindley's approximation method,” International Journal of Data Science, vol. 2, no. 4, pp. 352–368, 2017.
View at: Publisher Site | Google Scholar

Copyright

Copyright © 2022 Abdalla Rabie et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

PDF Download Citation

Download other formats

Order printed copies

Views

258

Downloads

352

Citations