A Parametric Empirical Bayes Model to Predict Software Reliability Growth

A new software reliability model based on the empirical Bayes estimate is developed. The number of failures estimated up to a given time is used in order to estimate the probability of failure appearance during the next time interval. Instead of a non homogeneous in time failure rate as it is usually used to model reliability growth, a failure rate depending non linearly on the previous number of failures is obtained from our model. The estimate is obtained from a mixed Poisson model where the mixing probability density function models the reliability growth. The model can be used either to simulate the cumulative failures curve or to estimate the time between failures. Data of a similar project can be used to estimate the parameters of a given project. Results of simulations and estimated mean time between failures comparing well with experimental data are also shown.