Dataset README
==============

Dataset name
------------
loan_v1_v5_complex_simulated

Short description
-----------------
This package contains a versioned semi-synthetic event log of a loan process
with five consecutive process versions. The dataset was generated to support
research on predictive process monitoring under structural process drift,
especially settings where the topology of a newly deployed process version is
known before representative execution logs of that version are available.

The dataset is intended as a controlled benchmark for topology-known structural
drift evaluation. It includes process evolution, new and modified activities,
routing changes, parallel branches, loops/rework, technical incidents, resource
variability, and cases that may cross process-version boundaries.

Files
-----
BPMN/
Directory containing BPMN process models for the five process versions:
loan_v1.bpmn, loan_v2_re.bpmn, loan_v3_re_pl.bpmn,
loan_v4_re_pl_cb.bpmn, and loan_v5_re_pl_cb.bpmn. These files provide
the deployed process topologies used for topology-known structural drift
evaluation.

loan_v1_v5_complex_simulated.xes
    Main event log in XES format.

loan_v1_v5_complex_simulated.summary.json
    Compact generation and dataset summary, including case counts, event
    counts, lifecycle distribution, activity frequencies, gateway branch
    counts, and resource utilization estimates.

loan_v1_v5_complex_simulated.dataset_stats.json
    Extended dataset statistics, including trace-length distributions,
    version carryover, cycle/rework statistics, node coverage, resource
    distributions, and operational-variance indicators.

Format
------
The event log is provided in XES format. Events follow a flattened XES-level
representation and include lifecycle transitions. The dataset intentionally
uses a basic transactional event schema rather than rich semantic task
descriptions.

Core event attributes
---------------------
The core event attributes are:

concept:name
    Activity name.

lifecycle:transition
    Lifecycle transition. The dataset contains assign, start, and complete
    events.

time:timestamp
    Event timestamp.

org:resource
    Resource identifier associated with the event.

Depending on the consuming tool, additional trace-level or event-level metadata
may be present. The JSON summary files should be used as the authoritative
source for aggregate statistics included in this package.

Generation summary
------------------
Process name: loan_v1_v5_simulated
Random seed: 4242
Simulation horizon: 2025-01-01T00:00:00+00:00 to 2025-06-01T00:00:00+00:00
Arrival process: Poisson, 1.54 cases/hour
Expected cases: 5580.96
Actual cases: 5559
Total events: 378679
Lifecycle events: 77937 assign, 150371 start, 150371 complete

Version-level case counts
-------------------------
v1: 1148 cases
v2: 1023 cases
v3: 1218 cases
v4: 1047 cases
v5: 1123 cases

High-level characteristics
--------------------------
Unique task nodes: 69
Mean number of completed task events per trace: 27.05
Maximum number of completed task events per trace: 95
Version carryover: 19.81% of cases finish in a later process version
Repeated activities: 6.55% of cases contain repetitions
Parallel execution: all cases include adjacent parallel interleaving
Incident-like cases: approximately 28.91% of cases include incidents
Resource variability: 87.85% of human task completions use a non-dominant
resource

Intended use
------------
The dataset is intended for:

1. predictive process monitoring experiments;
2. next-activity prediction under structural drift;
3. evaluation of fixed-vocabulary models under future-version candidate-space
   mismatch;
4. evaluation of topology-aware or topology-conditioned prediction methods;
5. reproducibility of the EOPKG structural-drift experiments.

The recommended experimental setting is to train on historical versions v1 and
v2 and evaluate future-version structural drift on v3, v4, and v5 without
training on logs from those future versions.

Known limitations
-----------------
This is a semi-synthetic dataset. It is designed as a controlled benchmark for
structural drift rather than as a direct sample of a specific real-world BPMS
deployment.

The data are intentionally limited to a flattened XES-level event schema. Full
execution marking, object-centric links, organizational hierarchies, textual
task descriptions, and rich resource profiles are not included.

Because the generated topology and the evaluation constraints share the same
design assumptions, topology masks may be cleaner than in noisy real-world
BPMS deployments.

Reproducibility materials
-------------------------
The full reproducibility package for the related study is expected to include:

1. BPMN/process-structure artifacts for process versions v1 to v5;
2. experiment configuration files;
3. MLflow run identifiers and metric exports;
4. aggregated result tables;
5. supplementary process-structure and drift-window materials.

Citation
--------
If you use this dataset, cite the related article and the dataset repository.
DOI: 10.5281/zenodo.20764041

License

This dataset is released under the Creative Commons Attribution 4.0
International license (CC BY 4.0). Users may share and adapt the dataset
provided that appropriate credit is given to the author and the dataset DOI is
cited.

Contact
-------
Serhii A. Korotenko
Private Higher Educational Establishment "European University"
Kyiv, Ukraine
e-mail: s.korotenko@e-u.edu.ua

