Statistical and Machine Learning Approach for Progressive Stress Accelerated Life Testing Under Unified Progressive Hybrid Censoring

ABSTRACT This article focuses on progressive‐stress accelerated life testing (ALT) under a unified progressive hybrid censoring scheme. The study assumes that the lifetimes of test units follow a Lomax distribution, with the scale parameter modeled using the cumulative exposure model and an inverse power law. The model parameters are estimated using both classical and Bayesian methods. The classical approach uses maximum likelihood estimation for point estimates and asymptotic confidence intervals. In contrast, Bayesian estimation employs informative and noninformative priors, with the Metropolis–Hastings algorithm handling complex posterior computations. Extensive simulations evaluate the performance of various progressive‐stress ALT, including simple and multiple ramp‐stress tests. We have applied the above methodology to real‐life data and demonstrated its applicability. Furthermore, we introduce machine learning techniques to analyze the reliability of the model. We have employed three machine learning methods, such as random forest regression, decision tree regression, and multilayer perceptron regression, to estimate reliability. A comparative analysis of all three methods using performance metrics has been discussed. Furthermore, we have compared the machine learning model with the statistical model for reliability estimation.