PREDICTING SOFTWARE-DEVELOPMENT ERRORS USING SOFTWARE COMPLEXITY METRICS

Various software metrics have been developed for the purpose of evaluating certain characteristics of the computer programming process. Attempts to use these software metric data have met with questionable success in program development scenarios. Similarly, there are many different models of software reliability. Each of these models fundamentally studies the computer program as a black box. The intrinsic assumption underlying each of the models is that the black box has one or more parameters that are uniquely determined for that particular box. Complexity metrics do provide substantial information on the distinguishing differences among the software systems whose reliability is being modeled and may be used in the determination of initial parameter estimates. Many of the existing complexity metrics measure common aspects of program complexity and thus have shared variance. In this paper, we develop predictive models that incorporate a functional relationship of program error measures with software complexity metrics and metrics based on factor analysis of empirical data. Specific techniques for assessing regession models are presented for analyzing these models. © 1990 IEEE