ISSUEFLOWMINER:Process and Sequential Pattern Mining of Issue Lifecycles in Open-Source Software Projects

V. International Perge Scientific Studies Congress, Antalya, Türkiye, 5 - 06 Eylül 2026, ss.71-89, (Tam Metin Bildiri)

Yayın Türü: Bildiri / Tam Metin Bildiri
Basıldığı Şehir: Antalya
Basıldığı Ülke: Türkiye
Sayfa Sayıları: ss.71-89
Marmara Üniversitesi Adresli: Evet

Özet

Software issue trackers record when issues are opened, labelled, discussed, linked to code, closed, and reopened. However, many studies examine these records mainly through counts and durations, which can obscure the order in which development activities occur. The main contribution of this study is an integrated, event-order-based framework that treats each issue as a process and connects three stages within the same analysis: initial resolution, reopening after closure, and the work performed after reopening. The framework was evaluated using a balanced stratified sample of 4,000 GitHub issues created between 2021 and 2024 in pandas, NumPy, SciPy, and scikit-learn. Complete issue timelines were collected, standardized into comparable activity types, and converted into event logs. After excluding seven records converted to discussions, the analytical dataset comprised 3,993 issues and 48,427 events. Process-mining techniques were used to visualize directly following activities and identify common end-to-end resolution paths. PrefixSpan was then used to discover recurring activity sequences before the first closure. Statistical models examined whether these sequences were associated with reopening within 180 days while accounting for repository, year, and multiple comparisons. The first reopening-to-reclosure cycle was also analyzed to characterize subsequent work. The ten most frequent resolution paths represented 94.08% of the issues. The median times to first response and initial closure were 3.50 and 123.89 hours, respectively. Among 277 candidate sequences, a label change followed by a code link was the sequence that retained statistical support after adjustment. This sequence occurred in 50.41% of the issues. The observed reopening rate was 1.84% among issues containing the sequence and 4.04% among the remaining issues. After repository and year differences were considered, the sequence was associated with approximately 60% lower odds of reopening (adjusted odds ratio = 0.403, 95% confidence interval: 0.271–0.600). Similar results were obtained using alternative reopening periods and sequence-frequency thresholds. Among 117 issues reopened within 180 days, the median duration of the first rework cycle was 184.77 hours. Comments and code links were observed in 80.34% and 52.99% of these cycles, respectively. The study provides a reproducible approach for moving beyond static issue attributes and examining software maintenance as an ordered process. The proposed framework may support process-aware comparisons of issue-management practices across open-source projects.

Keywords: Issue tracking, mining software repositories, process mining, sequential pattern mining, software maintenance