Identifying self-admitted technical debt in issue tracking systems using machine learning

Abstract Technical debt is a metaphor indicating sub-optimal solutions implemented for short-term benefits by sacrificing the long-term maintainability and evolvability of software. A special type technical explicitly admitted software engineers (e.g. using TODO comment); this called Self-Admitted Debt or SATD. Most work on automatically identifying SATD focuses source code comments. In addition to comments, issue tracking systems have shown be another rich SATD, but there are no approaches specifically in issues. paper, we first create training dataset collecting manually analyzing 4,200 issues (that break down 23,180 sections issues) from seven open-source projects (i.e., Camel, Chromium, Gerrit, Hadoop, HBase, Impala, Thrift) two popular Jira Google Monorail). We then propose optimize an approach machine learning. Our findings indicate that: 1) our outperforms baseline wide margin with regard F1-score; 2) transferring knowledge suitable datasets can improve predictive performance approach; 3) extracted keywords intuitive potentially types indicators SATD; 4) different less common compared same system; 5) small amount data needed achieve good accuracy.