External Code Quality Model and Cross-Validation of the Model
Lech Madeyski, Zbigniew Huzar · 2007
One goal of this paper is to empirically explore the relationships between existing object-oriented (OO) structural measures (both, class-level and package-level) and external code quality in the context of different development methods (solo/pair programming and test-first/testlast programming) and based on 122 projects. It appeared that external code quality can be better predicted based on OO measures (class-level CBO measure, package-level NOT measure), and size related LOCC measure than development methods. About 37 per cent of the external code quality variance can be explained by the model based on only aforementioned three measures. The second goal is to answer the question how accurate can these models be considering the unavoidable differences that may exist across projects and systems. This paper attempts to answer this question by means of cross-validation of the model. Twothirds of the projects (81 projects) were randomly picked to build our prediction model and the remaining one-third (41 projects) were used to verify the efficacy of the built model. Generalizability of the model, as currently captured by existing measures, seems to be limited as about 19 per cent of the variance can be explained by the model. Therefore, possible explanations and improvements are suggested.