PREDICTION AND CONTROL OF ADA SOFTWARE DEFECTS

A study of two Ada projects shows that software defects can be predicted in the development phase. Of all metrics tested, measures of package size proved to be the best predictors of defect densities. The prediction function derived suggests that defect density is very high for small Ada packages, decreases rapidly to reach a minimum for intermediate-sized packages, and then rises gradually as package sizes increase. During the design and coding phases, package sizes can be evaluated against this criterion, and such strategies as consolidation, further decomposition, redesigning, or recoding can be considered. Since the study found that packages with system integration and test (SI&T) defects are six times more likely to exhibit postdelivery defects as those without, a two-phased SI&T is suggested. © 1990.