Year: 2026
Type: preprint
Abstract: Predictive Process Monitoring (PPM) increasingly employs deep learning to predict future process behavior from historical event logs. However, existing approaches remain largely \emph{log-centric}, learning from system-enacted traces that represent individual executions rather than the behavioral semantics defined by the underlying process model. This semantic gap may cause predictive models to learn execution-specific artifacts instead of process-relevant dependencies.This paper proposes \emph{Model-Semantics-Aware Predictive Process Monitoring (MSA-PPM)}, a new paradigm that explicitly integrates model-defined process semantics into predictive learning. MSA-PPM introduces a \emph{Process Semantics Dictionary (PSD)} for representing parallel, exclusive-choice, loop, synchronization, temporal, resource, organizational, and constraint semantics. Guided by the PSD, conventional event traces are transformed into \emph{model-semantics-aware process traces} through semantic preprocessing prior to predictive learning, thereby separating semantic interpretation from predictive models and architectures.As an operational validation, we implement the parallel-semantics module of MSA-PPM. An exact Structured Information Control Net (SICN)-oriented process model is discovered using the $\rho$-Algorithm, and its parallel semantics are used to identify and eliminate directed phantom edges caused by arbitrary execution orders among parallel sibling activities. The semantically revised traces are evaluated for next-activity prediction using LSTM and Transformer models. Experimental results show absolute accuracy improvements of 5.63 and 4.90 percentage points, respectively, together with improvements in Precision, Recall, and F1-score. These results demonstrate that explicitly incorporating model-defined process semantics can improve predictive performance while providing an architecture-independent semantic foundation for PPM.
Source: Research Square
Language: English
Subfield: Management Information Systems
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7214923947