A Fuzzy Logic Based Set of Measures for Software Project Similarity: Validation and Possible Improvements

The software project similarity attribute has not yet been the subject of in-depth study, even though it is often used when estimating software development effort by analogy. Among the inadequacies identified (Shepperd et al.) in most of the proposed measures for the software project similarity attribute, the most critical is that they are used only when the software projects are described by numerical variables (interval, ratio or absolute scale). However, in practice, many factors which describe software projects, such as the experience of programmers and the complexity of modules, are measured in terms of an ordinal (or nominal) scale composed of qualifications such as ‘very low’, ‘low’ and ‘high’. To overcome this limitation, we propose a set of new measures for similarity when the software projects are described by categorical data. These measures are based on fuzzy logic: the categorical data are represented by fuzzy sets and the process of computing the various measures uses fuzzy reasoning. In this work, the proposed measures are validated by means of an axiomatic validation approach, using a set of axioms representing our intuition about the similarity attribute and verifying whether or not each measure contradicts any of the axioms. We also present in this paper the results of an empirical validation of our similarity measures, based on the COCOMO’81 database.