Leveraging Duration Pseudo-Embeddings in Multilevel LSTM and GCN Hypermodels for Outcome-Oriented PPM

TL;DR

This paper introduces a novel duration-aware pseudo-embedding approach integrated into multilevel LSTM and GCN hypermodels, significantly improving predictive accuracy, generalization, and interpretability in outcome-oriented predictive process monitoring tasks.

Contribution

It presents a dual input neural network with duration pseudo-embeddings for better temporal modeling in PPM, adaptable across different model architectures and datasets.

Findings

Duration pseudo-embeddings improve model generalization.

Models with pseudo-embeddings reduce complexity and enhance interpretability.

Experimental results show consistent performance gains across tasks.

Abstract

Existing deep learning models for Predictive Process Monitoring (PPM) struggle with temporal irregularities, particularly stochastic event durations and overlapping timestamps, limiting their adaptability across heterogeneous datasets. We propose a dual input neural network strategy that separates event and sequence attributes, using a duration-aware pseudo-embedding matrix to transform temporal importance into compact, learnable representations. This design is implemented across two baseline families: B-LSTM and B-GCN, and their duration-aware variants D-LSTM and D-GCN. All models incorporate self-tuned hypermodels for adaptive architecture selection. Experiments on balanced and imbalanced outcome prediction tasks show that duration pseudo-embedding inputs consistently improve generalization, reduce model complexity, and enhance interpretability. Our results demonstrate the benefits of explicit temporal encoding and provide a flexible design for robust, real-world PPM applications.