Empirical Bayes estimation of software failures

Abstract

The empirical Bayes estimator is applied to software failures production. The time between failures data registered up to a given time, are used in order to estimate the probability of failure appearance dur- ing the next interval time. This method is similar to the estimation of n-grams in natural language processing. A modi ed expression to the estimator usually used in language and speech processing is introduced in order to follow the failures production curve. Results of simulations comparing well with experimental data are also shown.