Empirically guided software development using metric-based classification trees

The identification of high-risk components early in the life cycle is addressed. A solution that casts this as a classification problem is examined. The proposed approach derives models of problematic components, based on their measurable attributes and those of their development processes. The models provide a basis for forecasting which components are likely to share the same high-risk properties, such as being error-prone or having a high development cost. Developers can use these classification techniques to localize the troublesome 20% of the system. The method for generating the models, called automatic generation of metric-based classification trees, uses metrics from previous releases or projects to identify components that are historically high-risk.