A composite complexity approach for software defect modelling

An approach is proposed to develop defect models for software components based on a categorical multivariate regression analysis. This modelling technique is useful when the software components are sufficiently small so that the assumption of a continuous normally distributed defect distribution is not valid. Library unit aggregations from five Ada projects are analysed to yield a composite complexity measure which is a function of both software complexity characteristics and development environment characteristics. The probabilities of various numbers of defects are derived from this composite complexity measure. The probability distributions are used to calculate subsystem level defects which are then compared to the actual defects.