197 American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) ISSN (Print) 2313-4410, ISSN (Online) 2313-4402 © Global Society of Scientific Research and Researchers http://asrjetsjournal.org/ Confidence Interval Estimation of the Conditional Reliability Function for Time Domain Data Lutfiah Ismail Al turk* Statistics Department, Faculty of Sciences, King Abdulaziz University, Jeddah, Kingdom of Saudi Arabia Email: lturk@kau.edu.sa Abstract The function of conditional reliability gives the probability of successfully implementing another operation following the successful implementation of a previous operation. The prediction of this function can help software developers in determining optimal release times. In this paper, the Maximum Likelihood Estimation (MLE) method is used to estimate the Non-Homogeneous Poisson Process Log-Logistic (NHPP LL) model’s parameters. The upper and the lower bounds of the parameters and conditional reliability function of time domain data are obtained. Real data application is conducted using the coefficient of multiple determination criteria and observed interval length to evaluate the performance of the NHPP LL model and the constructed confidence intervals, respectively. Our results encourage for more assessment of confidence intervals of other measures of reliability of the NHPP models. Keywords: NHPP log-logistic model; maximum likelihood estimation; confidence interval; conditional reliability function; observed interval length. 1. Introduction Software reliability is defined as the probability of failure–free operation of a computer program in a specified environment for a specified period of time [1], it received great attention due to its huge impact in our daily life [2,3]. Software reliability models based on Non-Homogeneous Poisson Process (NHPP) of time between failures class have been considered in the literature and validated as an accurate approach for estimating and predicting software reliability [4-7]. Hence, considering the Confidence Intervals (CIs) of software reliability can enhance the precision of the predictions for software testing. ------------------------------------------------------------------------ * Corresponding author. http://asrjetsjournal.org/ American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2019) Volume 52, No 1, pp 197-208 198 For software reliability data analysis, Yamada and Osaki [8] examined the Maximum Likelihood (ML) estimates using several SRGMs, they founded CI of the mean value function by the conventional NHPP method. Yin and Trivedi [9] obtained the confidence limits for the model parameters using the Bayesian method via implementing the estimation approach of Yamada and Osaki [8]. Huang [10] also followed the approach of Yamada and Osaki [13] to illustrate the CI of the mean value function graphically. This paper presents CI of the conditional reliability function of a NHPP model that assumes the time between two successive failures follow a Log-Logistic (LL) distribution. The LL distribution was first considered by Fisk [11], it is like the log-normal distribution, but with a little narrower peak and heavier tails. The LL distribution is among the class of survival time parametric models where the hazard function firstly increases and then decreases and at times can be hump-shaped, its mathematical simplicity and practicality has attracted many researchers in the field of survival analysis [12-14]. The paper layout of is as follows: Section 2 describes the conditional reliability function of the NHPP LL model. Section 3 discusses the parameter estimation and reliability prediction with confidence Intervals for the parameters and conditional reliability function of the NHPP LL model based on the times between failures data. Section 4 presents the analysis of three failure data sets, and Section 5 concludes the paper. 2. Conditional Reliability Function of a NHPP Model A NHPP model aim to estimate the expected number of faults experienced up to a certain point of time. If N(t) be the cumulative number of faults detected by the time t, F(t) is the distribution function and denote the expected number of faults that would be detected in a given infinite testing time, then the mean value function of a NHPP model is given by [15]: μ(ti; N0, Θ) = N0𝐹(ti; Θ), (1) where, N0 > 0 is the expected number of errors, F(ti; Θ) is the cumulative distribution of ti , i = (1, 2, … , n), Θ is its unknown parameters. Accordingly, the mean value function of the NHPP Log Logistic Model (NHPP LL model) is given below: μ(ti; N0, 𝜸, β) = N0𝛾𝑥 β 1+𝛾𝑥β , (2) where β > 0 is the shape parameter. and 𝛾 > 0 is positive scale parameter. The corresponding failure intensity function can be found by differentiating Eq. (2) as follows: η(ti; N0, 𝜸, β) = N𝛾𝛽ti β−1 (1+𝛾ti β) 2, (3) The conditional reliability function at time t of a NHPP model is exponential, given by: R(ti; N0, Θ|xn) = exp{−(μ(ti + xn; N0, Θ) − μ(ti; N0, Θ))}, where (4) https://en.wikipedia.org/wiki/Survival_analysis American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2019) Volume 52, No 1, pp 197-208 199 R(ti; N0, Θ|xn) is a monotone non-increasing function of ti; R(0; N0, Θ|xn) = 0 and R(∞;N0, Θ|xn) = 1. Consequently, the conditional reliability function of the NHPP LL model is given by: R(ti; N0, 𝛄, β|xn) = exp {−N0γ ( (ti+xn) β−xn β (1+γxn β )(1+γ(t+xn) β) )}. (5) More details about the NHPP LL model can be found in Al turk [16]. 3. Confidence Interval Estimation of the NHPP LL Model In this paper the MLE method will be applied to the time-interval between failures class of non- homogeneous Poisson process (NHPP) models. 3.1. Confidence interval estimation of the parameters Suppose that we have n observations represents the cumulative time to failures denoted by s1, s2, … , sn, then by considering Eqs. (2) and (3) the mean value and intensity functions of the NHPP LL model the log-likelihood function of N0, 𝛾, and 𝛽 can be written as: L(N0, 𝜸, β|S) = e −μ(ti;N0,𝜸,β)∏ 𝜂(ti; N0, 𝜸, β) n i=1 . (6) Taking the natural logarithm of Eq. (6) we obtain: lnL(N0, 𝜸, β|S) = −μ(ti; N0, 𝜸, β) + ∑ ln 𝜂(ti; N0, 𝜸, β) 𝑛 𝑖=1 = − 𝑁0𝛾𝑆𝑛 𝛽 1+𝛾𝑆𝑛 𝛽 + ∑ ln ( N0𝛾𝛽𝑆𝑖 𝛽−1 (1+𝛾𝑆 𝑖 𝛽 ) 2) 𝑛 𝑖=1 (7) = − 𝑁0𝛾𝑆𝑛 𝛽 1+𝛾𝑆𝑛 𝛽 + 𝑛 ln 𝛾 + 𝑛 ln𝛽 + 𝑛 lnN0 + 𝛽∑ 𝑙𝑛𝑠𝑖 𝑛 𝑖=1 − ∑ 𝑙𝑛𝑠𝑖 𝑛 𝑖=1 − 2∑ ln(1 + 𝛾𝑆𝑖 𝛽 )𝑛 𝑖=1 (8) Differentiating the above function with respect to 𝑁0, 𝛾, and 𝛽, we have { 𝜕lnL(N0, 𝜸, β|S) 𝜕𝑁0 = − 𝛾𝑆𝑛 𝛽 1+𝛾𝑆𝑛 𝛽 + 𝑛 𝑁0 . 𝜕lnL(N0, 𝜸, β|S) 𝜕𝛾 = 𝑛 𝛾 − 𝑁0𝑆𝑛 𝛽 1+𝛾𝑆𝑛 𝛽 + 2∑ 𝑆𝑖 𝛽 1+𝛾𝑆 𝑖 𝛽 𝑛 𝑖=1 . 𝜕lnL(N0, 𝜸, β|S) 𝜕𝛽 = 𝑛 𝛽 +∑ 𝑙𝑛𝑠𝑖 𝑛 𝑖=1 − 𝑁0𝛾𝑆𝑛 𝛽 𝑙𝑛𝑆𝑛 (1+𝜸𝑆𝑛 𝛽 ) 2 + 2∑ 𝛾𝑆𝑖 𝛽 𝑙𝑛𝑆𝑖 1+𝛾𝑆 𝑖 𝛽 𝑛 𝑖=1 . (9) The ML estimates can be obtained by setting the three expressions in Eq. (9) to zero as follows: American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2019) Volume 52, No 1, pp 197-208 200 { 𝑁0 = 𝑛 ( 1+𝜸𝑆𝑛 𝛽 𝛾𝑆𝑛 𝛽 ) . 𝑛 𝛾 − 𝑛 𝛾(1+𝛾𝑆𝑛 𝛽 ) + 2∑ 𝑆𝑖 𝛽 1+𝛾𝑆 𝑖 𝛽 𝑛 𝑖=1 = 0. 𝑛 𝛽 + ∑ 𝑙𝑛𝑠𝑖 − 𝑛𝑙𝑛𝑆𝑛 1+𝛾𝑆𝑛 𝛽 + 2∑ 𝛾𝑆𝑖 𝛽 𝑙𝑛𝑆𝑖 1+𝛾𝑆 𝑖 𝛽 𝑛 𝑖=1 = 0.𝑛 𝑖=1 (10) Due to the lack of explicit solutions to the second and third expressions of Eq. (10), we numerically find the estimates the parameters 𝛾 and 𝛽 then by substituting them in the first expression, N̂0 is obtained. To get the variance and covariance matrix for the estimated parameters, we first need to calculate the Fisher information matrix [17], which is: F = [ − ∂2 ln L ∂N0 2 − ∂2 ln L ∂N0 ∂γ − ∂2 ln L ∂N0 ∂β − ∂2 ln L ∂γ∂N0 − ∂2 ln L ∂γ2 − ∂2 ln L ∂γ∂β − ∂2 ln L ∂β∂N0 − ∂2 ln L ∂β ∂γ − ∂2 ln L ∂β2 ] , (11) where 𝜕2lnL(N0, 𝜸, β|S) 𝜕𝑁0 2 = −𝑛 𝑁0 2 , (12) 𝜕2lnL(N0, 𝜸, β|S) 𝜕𝛾2 = − 𝑛 𝛾2 + 2𝑁0𝑆𝑛 2𝛽 (1+𝛾𝑆𝑛 𝛽 ) 3 − 2∑ 𝑆𝑖 2𝛽 (1+𝛾𝑆𝑛 𝛽 ) 2 𝑛 𝑖=1 , (13) 𝜕2lnL(N0, 𝜸, β|S) 𝜕𝛽2 = − 𝑛 𝛽2 − 2𝑁0𝛾 2𝑆𝑛 2𝛽(𝑙𝑛𝑆𝑛) 2 (1+𝛾𝑆𝑛 𝛽 ) 3 + 2𝛾 ∑ 𝑆𝑖 𝛽(𝑙𝑛𝑆𝑖) 2 (1+𝛾𝑆 𝑖 𝛽 ) 2 𝑛 𝑖=1 , (14) 𝜕2lnL(N0, 𝜸, β|S) 𝜕𝑁0𝜕𝛾 = −𝑆𝑛 𝛽 1+𝛾𝑆𝑛 𝛽, (15) 𝜕2lnL(N0, 𝜸, β|S) 𝜕𝑁0𝜕𝛽 = − 𝛾𝑆𝑛 𝛽 𝑙𝑛𝑆𝑛 (1+𝛾𝑆𝑛 𝛽 ) 2, and (16) 𝜕2lnL(N0, 𝜸, β|S) 𝜕𝛾𝜕𝛽 = 𝑁0𝑆𝑛 𝛽 (α𝑆𝑛 𝛽 −1)𝑙𝑛𝑆𝑛 1+α𝑆𝑛 𝛽 + 2∑ 𝑆𝑖 𝛽 𝑙𝑛𝑆𝑖 (1+𝛾𝑆 𝑖 𝛽 ) 2 𝑛 𝑖=1 , (17) The asymptotic variance-covariance matrix is obtained by: Σ = F−1 American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2019) Volume 52, No 1, pp 197-208 201 = [ 𝑉𝑎𝑟(𝑁0) Cov(𝑁0, 𝛾) Cov(𝑁0, 𝛽) Cov(𝑁0, 𝛽) 𝑉𝑎𝑟(𝛾) Cov(𝛾, 𝛽) Cov(𝛽, 𝑁0) Cov(𝛽, 𝛾) 𝑉𝑎𝑟(𝛽) ], (18) So, the 100(1− 𝛼)% asymptotic confidence intervals for the parameters N0, 𝛾, and 𝛽of the NHPP LL model are given, respectively, by: (�̂�0 − 𝑍𝛼 2 √𝑉𝑎𝑟(�̂�0), �̂�0 + 𝑍𝛼 2 √𝑉𝑎𝑟(�̂�0) ), (19) (𝛾 − 𝑍𝛼 2 √𝑉𝑎𝑟(𝛾), 𝛾 + 𝑍𝛼 2 √𝑉𝑎𝑟(𝛾)), and (20) (�̂� − 𝑍𝛼 2 √𝑉𝑎𝑟(�̂�), �̂� + 𝑍𝛾 2 √𝑉𝑎𝑟(�̂�)), (21) where, 𝑍𝛼 2⁄ is the percentile of standard normal distribution with right-tail probability 𝛼 2⁄ , 𝑉𝑎𝑟(�̂�0), 𝑉𝑎𝑟(𝛾), and 𝑉𝑎𝑟(�̂�) are, respectively, the diagonal elements of the asymptotic variance and covariance matrix given by Eq. (18). 3.2. Confidence interval estimation of the conditional reliability function According to the invariance property of the ML estimators, the estimate of the conditional reliability of the NHPP LL model is obtained by: R̂(ti; N̂0, �̂�, β̂|xn) = exp {−N̂0γ̂ ( (ti+xn) β̂−xn β̂ (1+γ̂xn β̂ )(1+γ̂(t+xn) β̂) )} , (22) and its variance is defined as: 𝑉(�̂�) = ( 𝜕𝑅 𝜕𝑁0 ) 2 | 𝑁0=�̂�0 𝑉( �̂�0) + ( 𝜕𝑅 𝜕𝛾 ) 2 | 𝛾=�̂� 𝑉(𝛾) + ( 𝜕𝑅 𝜕𝛽 ) 2 | 𝛽=�̂� 𝑉(�̂�) + 2 ( 𝜕𝑅 𝜕𝛾 ) ( 𝜕𝑅 𝜕𝑁0 )| 𝛾=�̂�, 𝑁0=�̂�0 Cov (𝛾, �̂�0)+2 ( 𝜕𝑅 𝜕𝛾 ) ( 𝜕𝑅 𝜕𝑁0 )| 𝛾=�̂�, 𝑁0=�̂�0 Cov (�̂�, �̂�0) + 2 ( 𝜕𝑅 𝜕𝛾 ) ( 𝜕𝑅 𝜕𝛽 )| 𝛾=�̂�, 𝛽=�̂� Cov (𝛾, �̂�), (23) where 𝜕𝑅 𝜕𝑁0 = −𝛾 ( (𝑡𝑖+𝑥𝑛) 𝛽−𝑥𝑛 𝛽 (1+𝛾𝑥𝑛 𝛽 )(1+𝛾(𝑡+𝑥𝑛) 𝛽) ) 𝑒𝑥𝑝 {−N0𝛾 ( (𝑡+𝑥𝑛) 𝛽−𝑥𝑛 𝛽 (1+𝛾𝑥𝑛 𝛽 )(1+𝛾(𝑡+𝑥𝑛) 𝛽) )}, (24) American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2019) Volume 52, No 1, pp 197-208 202 𝜕𝑅 𝜕𝛾 = N0 ( (𝑡𝑖+𝑥𝑛) 𝛽−𝑥𝑛 𝛽 (1+𝛾𝑥𝑛 𝛽 )(1+𝛾(𝑡𝑖+𝑥𝑛) 𝛽) )( ((1+𝑥𝑛) 𝛽(1+2𝛾𝑥𝑛 𝛽 )+𝑥𝑛 𝛽 ) (1+𝛾𝑥𝑛 𝛽 )(1+𝛾(𝑡𝑖+𝑥𝑛) 𝛽) − 1) × 𝑒𝑥𝑝 {−N0𝛾 ( (𝑡𝑖+𝑥𝑛) 𝛽−𝑥𝑛 𝛽 (1+𝛾𝑥𝑛 𝛽 )(1+𝛾(𝑡𝑖+𝑥𝑛) 𝛽) )}, and (25) 𝜕𝑅 𝜕𝛽 = αN0 (1+𝛾𝑥𝑛 𝛽 )(1+𝛾(𝑡𝑖+𝑥𝑛) 𝛽) × ( 𝛾((𝑡𝑖+𝑥𝑛) 𝛽−𝑥𝑛 𝛽 )(𝑥𝑛 𝛽 (1+𝛾(1+𝑥𝑛) 𝛽)𝑙𝑛𝑥𝑛+(𝑡𝑖+𝑥𝑛) 𝛽(1+𝛾𝑥𝑛 𝛽 )𝑙𝑛(𝑡𝑖+𝑥𝑛)) (1+𝛾𝑥𝑛 𝛽 )(1+𝛾(𝑡𝑖+𝑥𝑛) 𝛽) − (𝑡𝑖 + 𝑥𝑛) 𝛽𝑙𝑛(𝑡𝑖 + 𝑥𝑛) − 𝑥𝑛 𝛽 𝑙𝑛𝑥𝑛 𝛽 ) × 𝑒𝑥𝑝 {−N0𝛾 ( (𝑡𝑖+𝑥𝑛) 𝛽−𝑥𝑛 𝛽 (1+𝛾𝑥𝑛 𝛽 )(1+𝛾(𝑡𝑖+𝑥𝑛) 𝛽) )}. (26) We employ the central limit theorem of the conditional reliability function and gets the asymptotic 100(1− 𝛼)% confidence bounds for the actual values as follows: (�̂� − 𝑍𝛼 2 √𝑉𝑎𝑟(�̂�), �̂� + 𝑍𝛾 2 √𝑉𝑎𝑟(�̂�)). (27) where, 𝑍𝛼 2⁄ is the percentile of standard normal distribution with right-tail probability 𝛼 2⁄ , �̂� is obtained from Eq. (22), and 𝑉𝑎𝑟(�̂�) is defined by Eq.(23). 4. Numerical Application A numerical application is illustrated in this section. The confidence interval estimation of the parameters and conditional reliability function of the NHPP LL model is investigated based on three real data sets. The NTDS data is obtained from Goel and Okumoto [17], it consists of 34 failures. The CSR2 and SYS2 data sets are from Lyu [3], the number of failures in these data sets are 129, and 86 failures, respectively. The three data sets are shown in Tables [1-3]. Table 1: NTDS data, 34 failures. 9 12 11 4 7 2 5 8 5 7 1 6 1 9 4 1 3 3 6 1 11 33 7 91 2 1 87 47 12 9 135 258 16 35 The coefficient of multiple determination R2 is used in our application to evaluate the model performance. Its formula is as follows [18]: R2 = 1 − ∑ (yi −m̂(ti)) 2𝑛 𝑖=1 ∑ (yi −∑ yk 𝑛⁄ 𝑛 𝑘=1 ) 2 𝑛 𝑖=1 . (28) https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4055612/table/tab1/ American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2019) Volume 52, No 1, pp 197-208 203 It takes the values from 0 to1. The larger value of R2 indicates better model performance. Though, the observed interval length is used to compare the CIs. The smaller the length, the better the confidence interval. Table 2: CSR2 data, 129 failures 760 758 303 6 22 14 42 4 84 15 221 14 15 41 1 153 409 54 24 44 180 397 19 145 36 54 1337 163 8 1 17 16 87 19 29 0 5 360 10 11 100 252 460 179 3 24 253 163 54 137 328 3 9 12 18 9 75 15 366 428 212 115 264 269 276 1 999 30 495 472 344 550 131 47 92 863 991 35 9549 249 607 83 614 352 673 4179 111 75 407 288 894 1314 845 55 409 36 15 1960 60 19 20 79 24 1737 7984 10 20 338 250 1682 212 287 56 4973 3500 59 98 2439 1812 6203 385 3500 4892 687 62 2796 3268 3845 76 Table 3: SYS2 data, 86 failures. 479 266 277 554 1034 249 693 597 117 170 117 1274 469 1174 693 1908 135 277 596 757 437 2230 437 340 405 535 277 363 522 613 277 1300 821 213 1620 1601 298 874 618 2640 5 149 1034 2441 460 565 1119 437 927 4462 714 181 1485 757 3154 2115 884 2037 1481 559 490 593 1769 85 2836 213 1866 490 1487 4322 1418 1023 5490 1520 3281 2716 2175 3505 725 1963 3979 1090 245 1194 994 3902 4.1. Numerical results The ML estimates and CIs at 95% significance level of the parameters N0, 𝛾, and 𝛽 are assessed using Eqs. (19), (20), and (21) corresponding to the last failure number of each data sets. For the comparison purpose the observed interval lengths of the CIs are computed as follows: 2𝑍𝛼 2⁄ [𝑉𝑎𝑟(N0̂)] 1 2⁄ , 2𝑍𝛼 2⁄ [𝑉𝑎𝑟(𝛾))] 1 2⁄ , 2𝑍𝛼 2⁄ [𝑉𝑎𝑟(�̂�)] 1 2⁄ . Also, to assess the model performance the coefficient of multiple determination criteria is computed for each of the selected data set, the results are summarized in Table 4. American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2019) Volume 52, No 1, pp 197-208 204 Table 4: Estimated Parameter Values of the NHPP LL Model and 95% Confidence Intervals. C.I. Lower C.I. Lower C.I. lower N̂0 γ̂ β̂ C.I. Upper C.I. Upper C.I. upper Data set Failure Number Observed Interval Length Observed Interval Length Observed Interval Length R2 Criteria 53.3639 0.0387 0.4484 NTDS data 34 53.614 0.0387 0.4947 0.7724 53.8641 0.0387 0.5411 0.5002 1e-04 0.0927 204.4341 0.024600 0.2936 CSR2 data 129 205.0794 0.024601 0.303 0.6393 205.7247 0.024603 0.3123 1.2907 3.32e-06 0.0188 199.9347 0.0199058 0.2704 SYS2 data 86 201.0775 0.0199064 0.2745 0.6052 202.2204 0.0199069 0.2786 2.2857 1.18e-06 0.0082 For the last three failure numbers, Table 5 illustrates the estimated conditional reliability of the NHPP LL model which is calculated using Eq. (22) and the corresponding 95% CIs with their observed interval lengths which are found using Eq. (27) and 2𝑍𝛼 2⁄ [𝑉𝑎𝑟(�̂�)] 1 2⁄ , respectively. Table 5: 95% Confidence Intervals of the conditional reliability function of the NHPP LL model. Data set Time to failure Estimated reliability at Time t C.I. lower C.I. upper Observed Interval Length 258 0.117 0.091 0.143 0.0536 NTDS data 16 0.113 0.087 0.139 0.0528 35 0.1047 0.0787 0.1307 0.052 3268 0.1592 0.1797 0.2002 0.0413 CSR2 Data 3845 0.1671 0.1466 0.1876 0.0412 76 0.1669 0.1464 0.1874 0.041 1194 0.168 0.1345 0.2016 0.0679 SYS2 data 994 0.1652 0.1316 0.1987 0.0675 3902 0.1545 0.121 0.1881 0.6052 American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2019) Volume 52, No 1, pp 197-208 205 Figures [1-3] demonstrate the estimated conditional reliability function and the corresponding 95% CIs for each selected data sets. Figure 1: Estimated reliability with 95% interval based on the NHPP LL model, NTDS data Figure 2: Estimated reliability with 95% interval based on the NHPP LL model, CSR2 Data 0 200 400 600 800 0 .0 0 .2 0 .4 0 .6 0 .8 1 .0 Time C o n d it io n a l R e li a b il it y F u n c ti o n 0 20000 40000 60000 80000 0 .0 0 .2 0 .4 0 .6 0 .8 1 .0 Time C o n d it io n a l R e li a b il it y F u n c ti o n American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2019) Volume 52, No 1, pp 197-208 206 Figure 3: Estimated reliability with 95% interval based on the NHPP LL model, SYS2 data From Tables 4 and 5, it can be noticed that the CIs estimated by the MLE method have small observed interval length which indicates the accuracy of these CIs. The assessment results, in Table 4, show that the estimator 𝛾 has the shortest observed interval length of the three selected data sets. Also, according to the model accuracy using the R2criteria the results in Table 4 show that the NHPP model fits best the NTDS data, then CSR2 Data and SYS2 data take the second and third rank, respectively. Regarding the lengths of the CIs presented in Table 5 it can be seen that as the number of detected failures increases narrower intervals of the conditional reliability function are obtained. The estimator R̂ has the shortest observed interval length for the CSR2 Data. 5. Conclusion It is essential to the software reliability measurement to obtain the confidence bounds for the reliability metrics at any future time t. The reliability function of a software system is an important metric for describing the system’s reliability. Our main contribution in this paper is to construct CI for the conditional reliability function of a NHPP model based on the LL distribution. CIs of the parameters and conditional reliability function of the NHPP LL model have been constructed based on the MLE method and evaluated via the observed interval length. The model performance has been checked using the R2 criteria. The application results demonstrate reasonable results for the CIs of the conditional reliability, which can help in improving the decision-making quality of software testing and debugging. Future research may find CIs for other reliability metrics of the NHPP models. 0e+00 2e+04 4e+04 6e+04 8e+04 1e+05 0 .0 0 .2 0 .4 0 .6 0 .8 1 .0 Time C o n d it io n a l R e li a b il it y F u n c ti o n American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2019) Volume 52, No 1, pp 197-208 207 References [1] Musa, J.D. (1998). Software Reliability Engineering McGraw –Hill. [2] Musa, J.D., Iannino A., and Okumoto K. (1987). Software Reliability Measurement, Prediction, Application, McGraw-Hill, New York. [3] Lyu, M. R. (1996). Handbook of software reliability engineering. NY: McGraw–Hill/IEE Computer Society Press. [4] Hossain S. A., Dahiya R. C. (1993). Estimating the Parameters of a Non-homogeneus Poisson-Process Model for Software Reliability, IEEE Transactions on Reliability, 42, pp. 604-612. [5] Gokhale S., Marinos P. N., and Trivedi K. S. (June 1996).“Important Milestones in Software Reliability Modeling”. In Proc. 8th Intl. Conference on Software Engineering and Knowledge Engineering (SEKE ’96), pages 345– 352, Lake Tahoe. [6] Pham H. and Zhang X. (2003). “NHPP software reliability and cost models with testing coverage,” European Journal of Operational Research, vol. 145, no. 2, pp. 443-454. [7] Almering, V. A., Genuchten, van, M. J. I. M., Sonnemans, P. J. M., & Cloudt, G. (2007). Using software reliability growth models in practice. IEEE Software, 24(6), pp. 82-88. DOI: 10.1109/MS.2007.182 [8] Yamada S. and Osaki S. (1985). “Software reliability growth modeling: Models and applications,” IEEE Transactions on Software Engineering, vol. 11, pp. 1431-1437. [9] Yin L. and Trivedi K. S. ( 1999 ). “Confidence interval estimation of nhpp based software reliability models,” in Proceedings 10th International Symposium on Software Reliability Engineering, pp. 6-11. [10] Huang C. Y. (2005) “Performance analysis of software reliability growth models with testing-effort and change-point,” Journal of Systems and Software, vol. 76: pp. 181-194. [11] Fisk PR. (1961). The graduation of income distributions. Econometrica, 29(2): pp. 171-185. http://dx.doi.org/10.2307/1909287 [12] Collett D. (1994). Modelling Survival Data in Medical Research. London: Chapman & Hall. [13] Gupta R.C., Akman O., Lvin S. (1999) A study of log-logistic model in survival analysis. Biometrical J, 41(4): pp. 431-443. [14] Bennett S. (1983). Log-logistic regression models for survival data. Appl Statist, 32(2): pp. 165-171. http://dx.doi.org/10.2307/2347295 [15] Lyu, M. R. (2002). Software Reliability Theory. Encyclopedia of Software Engineering, Wiley, pp. 1611-1630. [16] Lutfiah Ismail Al turk (Jan 2019). Characteristics and Application of the NHPP Log-Logistic Reliability Model, International Journal of Statistics and Probability; Vol. 8, No. 1, pp. 44-55, ISSN 1927-7032, E-ISSN 1927-7040. doi:10.5539/ijsp.v8n1p44 [17] Goel, A.L. and Okumoto, K. (1979). Time-Dependent Error-Detection Rate Model for Software Reliability and other Performance Measures, IEEE Trans. Reliability, R-28, 3, pp. 206-211. [18] Xie M., Yang B. (2003). A study of the effect of imperfect debugging on software development cost, IEEE Transactions on Software Engineering 29 (5), pp. 471–473. http://dx.doi.org/10.2307/1909287 http://dx.doi.org/10.2307/2347295 American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2019) Volume 52, No 1, pp 197-208 208 Author Information Lutfiah Ismail Al turk is currently working as associate professor of mathematical statistics in Statistics Department at Faculty of Sciences, King AbdulAziz University, Jeddah, Kingdom of Saudi Arabia. Lutfiah Ismail Al turk obtained her B.Sc degree in statistics and computer science from Faculty of Sciences, King AbdulAziz University in 1993 and M.Sc (mathematical statistics) degree from Statistics Department, Faculty of Sciences, King AbdulAziz University in 1999. She received her Ph.D in mathematical statistics from university of Surrey, UK in 2007. Her current research interests include software reliability modeling and statistical machine learning. Email: lturk@kau.edu.sa URL: http://lturk.kau.edu.sa 6. Address: P.O. Box 42713 Jeddah 21551. Kingdom of Saudi Arabia. mailto:lturk@kau.edu.sa http://lturk.kau.edu.sa/