Bug Prediction Metrics

[gmuecke]

Compute Number of potentially uncovered bugs, based on kLoC and Mutation Coverage. Based on the assumption that

• ~77% of all bugs can be covered by unit test (77% are achievable through 100% mutation coverage)
• industry statistics are between 15-50 bugs/kLoC before testing software, assuming 50 are for 0 tests
• microsoft application 10-20 bugs/kLoC before internal testing (assuming the 10 bugs are the remaining 23% not coverable by unit tests)

Threshold values and ratios should be coverable.

These metrics should give an indication of scale of number of bugs that are potentially hidden in the codebase and are left for test approaches other than unit test.

see also

• https://www.mayerdan.com/ruby/2012/11/11/bugs-per-line-of-code-ratio
• https://www.usenix.org/system/files/conference/osdi14/osdi14-paper-yuan.pdf
• https://people.cs.umass.edu/~rjust/publ/mutants_real_faults_fse_2014.pdf

In a future extension, the potential bugs could be comapared with actual bugs (i.e. from Jira)

[JoaoGFarias]

A misbehavior only exists in context: Somebody's expectation is not met.

"industry statistics" and "Microsoft application" apply to these contexts - their metrics don't apply to other's contexts, because their "somebody" are different people.

The proposed prediction model is an unscientific stretch of data that would mislead people.

[gmuecke]

agreed, I'm closing this