IssueLife: Synthetic issue trajectories, incomplete event logs and censored outcomes for remaining-time prediction benchmarks
Year: 2026
Type: dataset
Abstract: IssueLife is a synthetic dataset for evaluating remaining-time prediction when issue histories are incomplete and some outcomes are censored at an observation cutoff. Completion-duration distributions and issue-type proportions were calibrated using 309,792 resolved records from 151 projects in the public Itemlet dataset. Workflow transitions, time allocation among states and event-recording mechanisms were specified as simulation assumptions. The core release contains 48,000 trajectories derived from 8,000 underlying issue families across five random seeds, two workflow mechanisms and three scenarios. It includes 279,022 ground-truth events, 894,309 observed event records and 321,024 prediction snapshots at 24 and 168 hours after issue creation. Observation variants cover complete recording, random event loss, delayed recording and state-dependent event loss. Outcome tables distinguish follow-up observed before the cutoff, eventual completion and restricted remaining-time targets. Benchmark results include 2,936,136 individual predictions from seven comparators evaluated at 7-, 30- and 90-day remaining-time horizons. A separate twelve-condition sensitivity extension provides 3,622,752 additional prediction records. An independent descriptive comparison with the public Helpdesk event log reports substantial differences between real and synthetic traces. The repository provides the frozen dataset, executed Colab notebooks, generation and evaluation code, configurations, calibration grids, split assignments, data dictionaries, individual predictions, uncertainty estimates and integrity checks. These files support reproducible studies of observation loss, outcome censoring and feature availability. The dataset is intended for controlled methodological evaluation. Internal workflow histories, real-world missingness and censoring mechanisms have not been independently validated. The external comparison should not be interpreted as establishing realistic workflow behaviour.
Source: Mendeley Data
DOI: 10.17632/7j6n5s23px
Institution: Dayananda Sagar University
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225066053