Abstract
The improved adaptive progressive type-II censoring scheme has attracted considerable attention for its ability to effectively control test duration while still providing sufficient information for statistical inference from partially observed failure data. The existing literature on improved adaptive type-II progressive censoring schemes predominantly assumes predetermined removal schemes, in which the number of units removed at each failure time is fixed in advance. However, this assumption fails to account for the fact that removals may occur randomly in real-world applications. Accordingly, in this study, we consider an inference problem for lifetime data observed under an improved adaptive progressive type-II censoring scheme with random removals, where the number of units removed at each failure time follows a binomial distribution. The observed lifetime data are assumed to follow an exponential distribution for its analytical tractability and widespread use in reliability and survival analysis. Under this censoring scheme, point and interval estimates are obtained for the rate parameter and the reliability function of the exponential distribution, as well as for the binomial removal parameter. Point estimators are derived using maximum likelihood and Bayesian approaches, while interval estimates are constructed using asymptotic confidence intervals based on the maximum likelihood estimators and highest posterior density credible intervals derived from the Bayesian framework. The performance of the proposed estimators is evaluated through an extensive Monte Carlo simulation study using average estimates, mean squared errors, and average interval lengths. The results indicate that the Bayesian estimators consistently outperform the corresponding maximum likelihood estimators, yielding smaller mean squared errors and shorter intervals across all considered simulation settings. In addition, a real data application is presented for illustrative purposes. Overall, the present study provides a computationally efficient and flexible inferential procedure for the exponential distribution under the improved adaptive progressive type-II censoring scheme with binomial removals, which reflects the randomness of removals encountered in practical life-testing experiments. Furthermore, the results of this study indicate that Bayesian estimators based on prior information perform better than their maximum likelihood counterparts. In this context, the findings of the present study may serve as a useful reference for researchers and practitioners working on applications in reliability engineering, survival analysis, and related fields where censoring is commonly encountered.