Experimental evaluation of software reliability growth models

An experimental evaluation is presented of SRGMs (software reliability growth models). The experimental data sets were collected from compiler construction projects completed by five university students. The SRGMs studied are the exponential model, the hyperexponential model, and S-shaped models. It is shown that the S-shaped models are superior to the exponential model in both the accuracy of estimation and the goodness of fit (as determined by the Kolmogorov-Smirnov test). It is also shown that it is possible to estimate accurately residual faults from a subset of the test results. An estimation method is proposed for the hyperexponential model. It is based on the observation that the start time for testing is different for different program modules. It is shown that this method improves the goodness of fit significantly.<>