Empirical Validation of Object-Oriented Product Measures

Software costs are rising, as are problems with software quality and delivering software on schedule and within the development budget. As a result, more companies have begun to develop software using an object-oriented development methodology because advocates claim that such techniques can help reduce costs and increase software quality. At the same time managers have begun to argue that empirical evidence is needed to convince them of the benefits of such technologies. Many different avenues of empirical research are available, but an area which has recently been gathering momentum is software metrics; in particular software product metrics. An important advantage offered by product metrics, if available early in the design process, is that they allow managers and developers to better understand, control and improve their software development practices. The work for this thesis will involve the following: (i) build upon the literature review of OO product metrics conducted by Wuest, (ii) select the most important, i.e. those metrics which have sensible empirical models, and interesting, i.e., those metrics which are available early in the development process, product metrics with respect to cohesion, coupling and inheritance, and (iii) investigate (by using an existing data set of C++ programs and collected fault data for them) through the techniques of factor analysis, logistic regression, and outlier analysis. The results should help managers and developers to evaluate designs early with respect to their quality, i.e., in this case fault proneness, giving more control over the development process, e.g., to provide early feedback to evaluate the quality of alternative designs, for where to focus design or code inspections, or for the allocation of resources with respect to testing activities. In addition this should help save money over the course of the software development process