Machine Discovery of Static Software Reuse Potential Metrics

This paper reports a study to identify static software reuse potential metrics that can be used to classify C source code into reusable and non-reusable classes. The techniques used exploit a decision tree inductive machine learning and rough sets theory. The results we obtained show that the former technique, as implemented by C4.5, produces a much more accurate set of classification rules than the latter technique, as implemented by DataLogic/R. The C4.5 rules are also plausible as they support current understanding of how software metrics can be used to measure software reuse potental. keywords: inductive concept learning, machine learning, rough sets, software reuse track: intelligent system technologies (machine learning)