Completely monotone regression estimates of software failure rates

摘要

A new method for estimating the present failure rate of a program is presented. A crude nonparametric estimate of the failure rate function is obtained from past failure times. This estimate is then smoothed by fitting a completely monotonic function, which is the solution of a quadratic programming problem. The value of the smoothed function at present time is used as the estimate of present failure rate. A Monte Carlo study gives an indication of how well this method works.