Process routing and decision driver discovery from relational databases without prior event logs

Publication Type

Conference Proceeding Article

Publication Date

1-2026

Abstract

Classical process mining derives process models from event logs, but in many operational systems such logs are unavailable or incomplete. Instead, process execution is recorded implicitly in relational databases, making it difficult to determine how process instances progress between tables or why alternative process paths occur. This paper introduces Process ROuting Discovery from relational data (PROD), a framework that infers process routing behaviour directly from relational data-bases without requiring event logs. PROD identifies candidate successor relations between tables and assigns row-level routing across linked rows. It then learns routing decisions and their associated driving attributes at branching points, and constructs a routing graph whose branching behaviour is explained using SHAP-based attributions and expressed via a Large Language Model (LLM). We evaluate PROD on six datasets spanning synthetic, benchmark, and industrial relational databases. Results show that process behaviour and its driving attributes can be recovered, achieving routing accuracy ranging from 0.79–0.96 and Structural F1-scores of 0.83–0.95, while exhibiting gradual performance degradation as row-level data corruption increases. These findings suggest that PROD enables practical process analysis directly from relational data, broadening the scope of process analysis beyond traditional event-log-based settings.

Keywords

Process Analysis, Routing Analysis, Explainable Process

Discipline

Artificial Intelligence and Robotics | Databases and Information Systems

Research Areas

Intelligent Systems and Optimization

Publication

24th International Conference on Business Process Management (BPM)

First Page

485

Last Page

502

ISBN

9783032377760

Identifier

10.1007/978-3-032-37777-7_29

Publisher

Springer

City or Country

Cham

Additional URL

https://doi.org/10.1007/978-3-032-37777-7_29

This document is currently not available here.

Share

COinS