No abstract

We present an approach to automatically discovering explicit rules for software process evaluation from evaluation histories. Each rule is a conjunction of a subset of attributes in a process execution, characterizing why the execution is normal or anomalous. The discovered rules can be used for stakeholder as expertise to avoid mistakes in the future, thus improving software process quality; it can also be used to compose a classifier to automatically evaluate future process execution. We formulate this problem as a contrasting itemset mining task, and employ the branch-and-bound technique to speed up mining by pruning search space. We have applied the proposed approach to four real industrial projects in a commercial bank. Our empirical studies show that the discovered rules can precisely pinpoint the cause of all anomalous executions, and the classifier built on the rules is able to accurately classify unknown process executions into the normal or anomalous class.

No abstract

No abstract

No abstract