Cyclomatic complexity as a utility for predicting software faults

Abstract

The cyclomatic complexity metric provides a means of quantifying intra-modular software complexity, and its utility has been suggested in the software development and testing process. In this thesis, an empirical analysis was undertaken to examine the relationship between the cyclomatic complexity and the incidence of faults in a series of eight relatively large (from 1200 to 2400 LOC) complex programs. Each of these programs was developed form a single program specification and subsequently subjected to rigorous unit level testing. A comparison was also made between the relationship of cyclomatic complexity to faults and the relationship of Lines of Codes (LOC) to faults. The results of this study support a relationship between the cyclomatic complexity and the incidence of faults. Further, a relationship between LOC and faults is demonstrated. It could not be shown that there exists a stronger relationship between cyclomatic complexity and faults than LOC and faults.