摘要：Early bug detection reduces the cost of software maintenance, but previous works have not utilized requirement traceability links (RTLs) as predictors for bugs. To discuss how to use RTLs to predict the number of bugs, we propose an RTL recovery approach classification based on the ease of the recovery process. We investigate the relationship using data from industrial software. Classes related to more RTLs tend to have more bugs. The classification provides better correlations, and including RTLs in the bug prediction model does not affect the performance. Some class files with no and low RTLs also have bugs; we hypothesize that this occurs because the actual RTL is missing or not established, which is supported by the observation that bugs in these classes are highly correlated with the maximum cyclomatic complexity.
