Software metrics through fault data from empirical evaluation using verification & validation tools

Traditional software metrics have been highly criticized for their inadequate measuring of software properties. Some metrics have been found to be based upon poor theoretical foundations (McCabe) while others are just too simplistic (Lines of Code). This thesis introduces a new methodology for creating software metrics. By defining metrics based upon where errors are typically discovered in code, with a particular emphasis on a composition of data and control flow, a new metric is proposed which measures the propensity of error prone areas in code. An empirical evaluation on software defects is used to identify error prone areas of source code. This evaluation uncovered characteristics shared by high error density code areas. Based on these common characteristics, a new software metric is defined. The new metric is then evaluated using a popular set of metric properties and validated using a statistical correlation.