A four-project empirical study of software error prediction models

Rome Air Development Center (RADC) has initiated an in-house study to validate several mathematical models for predicting the reliability and error content of a software package against error data extracted from four large Department of Defense software development projects. This paper will describe some of the results of this empirical study for three such models, the Jelinski-Moranda, Schick-Wolverton and a modified Schick-Wolverton. Model predictions will be compared on a functional subsystem basis, and on a total system versus summed subsystem and criticality basis. Model predictions will also be compared on a daily vs. weekly basis for defining model time intervals, and the question of when to begin applying these models will be addressed. Finally, general conclusions will be drawn as to model applicability.