Using Self-Organizing Maps to Analyze Object-Oriented Measures

t In this study, we present self-organizing maps and discuss their role in the analysis and visualization of software modules in the space of software measures. We reveal how self-organizing maps create a user-friendly and interactive visualization tool that helps user/software designer inspect various alternatives and get a thorough insight into the structure of the clusters of the software modules and the related metrics (software measures). We show how using self-organizing maps we can grow clusters in a dynamic fashion thus explicitly capture relationships between the software measures and quantify these dependencies for larger and less homogeneous clusters of software modules. The experimental environment exploited in this study relies on software measures coming from 10 large public domain systems, 5 Java and 5 C++ systems.