A Neural Network Paradigm for Characterizing Reusable Software

Deriving a measure for the reusability of software components has proven to be a challenging task. As with much human related assessment, the transformation of intuitive human evaluation into a concise polynomial representation is problematic, given the holistic nature of that human intuition. Many metrics exist intuitively without mathematical models. We describe here an alternative approach to the assessment of component reusability based upon the training of neural networks to mimic a set of human evaluators. We show that a neural approach is not only feasible, but can achieve good results without requiring inputs other than those readily available with metrics evaluation packages.