Harmonic mean estimates for recapture debugging
Abstract
In this thesis we examine the problem of estimating the number of errors (bugs) in a reliability system using the recapture debugging model suggested by Nayak (1988). The reliability system contains a certain number N of errors. Each causes system failures independently of the others. The times between failures for any bug are assumed to be independent exponential random variables with a parameter X common to all bugs. We assume the system is observed for a fixed length of time. The maximum likelihood estimate of N was considered by Nayak. We derive the profile likelihood interval for N, and consider as a point estimate the harmonic mean of the endpoints. This estimate was used for the Jelinski-Moranda model by Joe and Reid (1985). The exact probability distribution of the harmonic mean estimator is computed. A generalization of the harmonic mean estimate, called the weighted harmonic mean estimate is proposed as a further improvement. A comparison is drawn between this estimator and the maximum likelihood estimator, using their computed distributions for various values of N.
Type
Thesis, MSc Master of Science
Collections
Items in the St Andrews Research Repository are protected by copyright, with all rights reserved, unless otherwise indicated.