Performance comparison of software complexity metrics in an open source project

Abstract

Software complexity measures are essential aspects of software engineering. Relatively few studies have been conducted to compare the performance of different complexity metrics. This paper describes an experimental investigation, which compares the performance of three different software complexity metrics; McCabe's cyclomatic complexity, Halstead's complexity measures and Douce's spatial complexity, by using data from an Open Source project Eclipse JDT. The results of this investigation indicate that in different situation these complexity metrics show different performance. However, Halstead's effort measure and Douce's spatial complexity are highly correlated, showing bigger correlation coefficient values. This leads us to suggest that because Halstead's complexity measure is more mature and has better supporting tools, it may be a good idea to replace Douce's spatial complexity metric with Halstead's effort measure in practice.