Mining Frequent Patterns from Software Defect Repositories for Black-Box Testing

Abstract

Software defects are usually detected by inspection,black-box testing or white-box testing. Current software defect mining work focuses on mining frequent patterns without distinguishing these different kinds of defects, and mining with respect to defect type can only give limited guidance on software development due to overly broad classification of defect type. In this paper, we present four kinds of frequent patterns from defects detected by black-box testing (called black-box defect) based on a kind of detailed classification named ODC-BD (Orthogonal Defect Classification for Blackbox Defect). The frequent patterns include the top 10 conditions (data or operation) which most easily result in defects or severe defects, the top 10 defect phenomena which most frequently occur and have a great impact on users, association rules between function modules and defect types. We aim to help project managers, black-box testers and developers improve the efficiency of software defect detection and analysis using these frequent patterns. Our study is based on 5023 defect reports from 56 large industrial projects and 2 open source projects.