A1802
Title: Optimal design of Type-I censoring experiments via Shannon information gain
Authors: Hon Yiu So - Oakland University (United States) [presenting]
Sukanya Das - Oakland University (United States)
Abstract: Type-I censoring is widely used in life-testing and reliability studies to control experimental duration and cost, but the choice of censoring time and sample size critically affects the information obtained from the data. A Bayesian framework for optimal design of Type-I censoring experiments based on Shannon information gain, defined as the expected Kullback-Leibler divergence from prior to posterior, is presented. A computational approach that combines augmented probability simulation with a Metropolis-Hastings algorithm is developed to explore the design space of censoring time and sample size. A monotone spline smoothing technique is further employed to estimate a stable and interpretable information surface. Theoretical results establish that the expected information gain is non-decreasing in both censoring time and sample size, with diminishing marginal returns, providing practical guidance for design selection. Simulation studies and real data analysis demonstrate how the proposed method identifies efficient censoring schemes under budget constraints. While motivated by reliability applications, the framework applies broadly to censored lifetime experiments, including settings where only limited or indirect failure information is observed.
