Narrow Results

Software measures (metrics) provide software engineers with an important means of quantifying essential features of software products and software processes such as software reliability, maintenance, reusability and alike. Software measures interact between themselves. Some of them may be deemed redundant. Software measures are used to construct detailed prediction models. The objective of this study is to pursue an association analysis of software measures by revealing dependencies (associations) between them. More specifically, the introduced association analysis is carried out at the local level by studying dependencies between information granules of the software measures. This approach is contrasted with a global level such as e.g., regression analysis. We discuss the role of information granules as meaningful conceptual entities that facilitate analysis and give rise to a user-friendly, highly transparent environment.

In this study, we highlight some fundamental issues of knowledge management and cast them in the setting of Granular Computing (GrC). We show how its formal constructs — information granules are instrumental in knowledge representation and specification of its level of abstraction.

This paper proposes a case-based classifier using a new approach that integrates rule-based and case-based reasoning approaches for enhanced accuracy. The rule-based reasoning component uses rules generated from a concept lattice of training data, binarized using fuzzy sets. These binarized data are stored as cases in the case-based classification component. The case-based component complements the rule-based component to enhance classification accuracy. Moreover, we designed the case-based component with an embedded similarity measure that uses a vector model for concept approximations. Thus, this design makes it possible to generate high quality rules and classify unseen new cases. In addition, the ability to build a knowledge base in lattice form is important for discovering hierarchical patterns, incrementing or updating the existing knowledge base, and inducing rules with our rule learning algorithm. The novel methodology was implemented and evaluated with benchmark datasets from the UCI repository and historic rubber prices in Thailand, demonstrating improvements in accuracy of classification calls. The results from the fact their several hierarchical datasets are very promising, with improved classification performance over prior reported methods.