Estimation of the Total Number of Software Failures from Test Data and Code Coverage: A Bayesian Approach
Harald A. Stieber, Linghuan Hu, W. Eric Wong · 2017
Total number of failures of a software system can help practitioners to have a better understanding of the software quality. In this paper, we propose a model to predict the total number of software failures in a software system by analyzing the failure data from testing using models based on Zipf's law together with the information on code coverage. Failure data and code coverage are combined in a Bayesian way. The methodology is applied to real world failure data to validate its predictability. The predictive accuracy of our model is also evaluated with different methods. The results of our experiment show that our proposed model can provide a very good estimation of the total number of failures. The estimation is stable from a very early point on.