Estimating the Total Number of Software Faults Reliability Models and Mutation Testing a Bayesian Approach

The methodology presented in this paper uses failure data from the software test and combines them with the results from mutation testing. New models based on Zipf´s law are used - together with a suitable method for parameter estimation - to get interval estimates for the total number of faults. The new models have been applied to several well-known data sets and some data sets from the author. In all cases the results were excellent. The new models and the parameter estimation method are especially suitable for grouped data (interval domain data). In a next step the confidence interval above is combined - in a Bayesian way - with the results from mutation testing. Now you get a new (and better) confidence interval for the total number of faults and an assessment of the goodness of the software test.