An analysis of faulty and fault-free C++ classes using an object-oriented metrics suite

Abstract

In this paper, we analyze a dataset freely provided through the PROMISE repository containing data on object-oriented (OO) class features and associated faults. We used a set of metrics provided with the dataset, based loosely on the Chidamber & Kemerer (C&K) metrics for our analysis; in particular, we compared and contrasted the characteristics of classes containing faults with those containing zero faults as a mechanism for establishing their causes and a hypothesis-based approach was adopted to this end. Several key results emerged from our analysis. Firstly, coupling seems to be a key factor influencing fault-proneness; the likelihood of at least one fault is greater when there is relatively high coupling. Secondly, class size 'does matter' - the more methods in a class, the more faults the class tends to contain. Finally, cross-correlation of the five metrics revealed an interesting trait related to inheritance and a previous study into C++ friends. © Springer Science+Business Media B.V. 2008.