Mining Bugzilla datasets with new increasing failure rate software reliability models

A data mining of several Bugzilla datasets using Software Reliability models is presented. We analyzed Bugzilla reports from the Xfce, Firefox, Eclipse and Tomcat projects for a long period of time of several thousand of days. In all the cases, an increasing failure rate have been found. Increasing failure rates are usually modeled by the S-shaped model in the literature. We propose to use some models where the failure rate depends not just on time, but also on the number of failures previously reported. These models are little modifications of the Yule and Polya stochastic processes. The comparison of the actual data with results from simulations is presented. Those results show that many Open Source bugs datasets can be fit by the proposed models.