MDSA-OPS

Code for the paper submitted to CAiSE 2026 on multi-source drift-aware online process simulation.

---

Installation

1. Create a Python environment

We recommend using Conda:

conda create -n obps python=3.11
conda activate obps

2. Install Python dependencies

pip install -r requirements.txt

Data Preparation

Place your event streams under the streams/ directory and specify the corresponding data_path in the configuration file.

• The repository currently provides the event streams used in our experiments.
• You can add your own streams by placing them in streams/ and updating the config accordingly.

Configuration

Model and experiment settings are specified in YAML files under the configs/ directory.

Key parameters include:

• data_path: path to the input event stream.
• process_fitness_threshold: fitness threshold for the process model N.
• process_error_threshold: error threshold for the branch model $P_B$.
• arrival_error_threshold: error threshold for the arrival model A.
• res_error_threshold: error threshold for the resource model R.
• wt_error_threshold: error threshold for the waiting-time model W.
• et_error_threshold: error threshold for the execution-time model E.

Running

Once the configuration file is prepared, run:

python OnlineSimulation.py

By default, the script will:

1. Load the YAML configuration from configs/.
2. Read the event stream from data_path (under streams/).
3. Perform multi-source drift-aware online process simulation and output evaluation results.

Model Architecture

The overall architecture of MDSA-OPS is illustrated in the following figure.

Experimental Results

The overall performance comparison between baseline methods and OURS is summarized below.
Lower values indicate better alignment between simulated and real behavior (smaller distance).
(Control-flow: NGD, CFLD; Temporal: AED, CAD, CED, RED, CTD; Workforce: CWD.)

Event Log	Method	NGD	CFLD	AED	CAD	CED	RED	CTD	CWD
BPIC12W	SIMOD[1]	0.250	0.274	1251.559	1314.318	5.079	62.610	131.025	4.426
	AgentSim[2]	0.320	0.343	1500.402	1654.119	3.846	162.442	280.662	3.657
	OBPS[3]	0.234	0.327	274.896	114.320	2.513	66.424	83.840	3.285
	OURS	0.109	0.100	114.423	35.264	1.261	45.434	37.983	1.521
BPIC17W	SIMOD	0.257	0.184	860.145	861.299	4.380	36.702	73.045	4.056
	AgentSim	0.147	0.139	1337.698	1360.753	2.417	67.525	101.322	2.732
	OBPS	0.356	0.418	314.311	381.286	1.427	54.129	90.305	2.042
	OURS	0.140	0.126	256.534	58.013	1.659	32.232	57.812	1.765
ACR	SIMOD	0.250	0.164	856.832	856.046	3.272	83.640	240.823	3.308
	AgentSim	0.246	0.186	391.468	353.958	5.268	118.395	334.011	4.846
	OBPS	0.656	0.413	151.373	67.808	1.393	103.558	223.030	1.478
	OURS	0.164	0.100	107.171	21.125	1.841	46.720	138.209	1.901
Production	SIMOD	0.669	0.567	1822.346	1786.794	4.456	59.573	137.325	4.481
	AgentSim	0.615	0.566	512.280	501.273	3.680	83.232	159.661	3.757
	OBPS	0.729	0.627	192.750	23.005	1.706	45.385	172.162	1.759
	OURS	0.455	0.439	156.432	40.886	0.940	52.953	69.772	0.814

The simulation logs generated by our method are provided in the log_results/ directory.

References

[1] Camargo M, Dumas M, González-Rojas O. Automated discovery of business process simulation models from event logs. Decision Support Systems, 2020, 134: 113284.
Code: https://github.com/AutomatedProcessImprovement/Simod

[2] Kirchdorfer L, Blümel R, Kampik T, et al. Agentsimulator: An agent-based approach for data-driven business process simulation. 2024 6th International Conference on Process Mining (ICPM). IEEE, 2024: 97-104.
Code: https://github.com/lukaskirchdorfer/AgentSimulator

[3] Vinci F, Park G, Van Der Aalst W M P, et al. Online Discovery of Simulation Models for Evolving Business Processes. International Conference on Business Process Management. Cham: Springer Nature Switzerland, 2025: 451-468.
Code: https://github.com/franvinci/ProcessSimulationTool