Empirical validation of class diagram complexity metrics

One of the principal objectives of software engineering is to improve the quality of software products. It is widely recognised that the quality assurance of software products must be guaranteed from the early phases of development. As a key artifact produced in the early development of object-oriented (OO) information systems (OOISs), class diagram quality has a great impact on the quality of the software product which is finally delivered. Hence, class diagram quality is a crucial issue that must be evaluated (and improved if necessary) in order to obtain quality OOISs, which is the main concern of present-day software development organisations. After a thorough review of the existing OO measures that are applicable to class diagrams at a high-level design stage, M. Genero et al. (2000) presented in a set of metrics for the structural complexity of class diagrams built using the Unified Modelling Language (UML). We focus on class diagram structural complexity, an internal quality attribute which we believe could be closely correlated with one of the most critical external quality attributes, such as class diagram maintainability. Since the main goal of this paper is the empirical validation of those metrics, we present two controlled experiments carried out to corroborate if those metrics are closed to class diagram maintainability and thus could be used as early maintainability indicators. Based on data collected in the experiments, we build a prediction model for class diagram maintainability using a method for the induction of fuzzy rules