Empirical Evaluation of Bug Proneness Index Algorithm

Researchers have devised and implemented different bug prediction approaches that use different metrics to predict bugs in software modules. However, the focus of research has been on proposing new approaches/models to predict bugs rather than on validating performance of existing approaches. In this paper, the authors evaluate and validate the findings of an algorithm that predicts the bug proneness index (bug score) of the software classes/modules. The algorithm uses normalized marginal R square values of software metrics as weights to the normalized metrics to compute bug proneness index (bug score). The experiment was performed on Eclipse JDT Core and reports significant improvements in F-measure of their algorithm as compared to the multiple linear regression. The authors found that there was no improvement in F-measure of evaluated algorithm compared to multiple linear regression. The use of marginal R square values as weights to the metrics in linear functions in the evaluated model instead of regression coefficients had no performance boost compared to the multiple linear regression.