Decomposition Methods

Abstract

Several types of data, e.g., software product attributes, process attributes and personnel attributes (see Section 2.1.4), are used in the empirical regression methods to estimate the number of software defects. In the dynamic methods, the type of input data is not consistent with that of output data. If the principle of data type conservation works well in software reliability modeling, it should not be surprising that the empirical methods offer more qualitative than quantitative information (the methods partially comply with the principle), whereas the validity of the dynamic methods should be doubted (the methods violate the principle). It seems that the capture-recapture methods comply with the principle of data type conservation. A major problem with the capture-recapture methods is that software defect seeding has to be applied. A method which can avoid this problem while complying with the principle of data type conservation should be more desirable. In this chapter we develop such a method and try to justify its validity. The key idea of this method is to decompose the software into two parts and identify the location of each observed defect (this is why we call the method decomposition method) [2]. We note that few capture-recapture methods or other methods, if any, identify the location of observed defects, although this kind of information is available and can be easily collected.