Complexity Metrics Significance for Defects: An Empirical View

Abstract

Software Complexity often seems to be correlated with the defects and this makes difficult to select appropriate complexity metrics that would be effective indicators of defects. The aim of this work is to analyze the relationship of different complexity metrics with the defects for three categories of software projects i.e. large, medium and small. We analyzed 18 complexity metrics and defects from 27,734 software modules of 38 software projects categorized in large, medium and small. In all categories of projects we do not find any strong positive correlation between complexity metrics and defects. However, we cluster the complexity metric values and defects in three categories as high, medium and low. Consequently we observe that for some complexity metrics high complexity results in higher defects. We called these metrics as effective indicators of defects. In the small category of projects we found LCOM as effective indicator, in the medium category of project we found WMC, CBO, RFC, CA, CE, NPM, DAM, MOA, IC, Avg CC as effective indicators of defects and for a large category of projects we found WMC, CBO, RFC, CA, NPM, AMC, Avg CC as effective indicators of defects. The difference shows that complexity metrics relation to defects also varies with the size of projects.