---
title: "LSTM Networks for Data-Aware Remaining Time Prediction of Business Process Instances"
type: source
tags: [ppm, lstm, remaining-time, data-aware, foundational]
authors: [Navarin Nicolò; Vincenzi Beatrice; Polato Mirko; Sperduti Alessandro]
year: 2017
venue: "IEEE Symposium Series on Computational Intelligence (SSCI) 2017"
kind: paper
raw_path: "raw/Predictive process monitoring/LSTM Networks for Data-Aware Remaining Time Prediction of Business Process Instances.pdf"
created: 2026-04-13
updated: 2026-04-13
key_claims:
  - Additional case information (data attributes) materially improves remaining-time prediction quality.
  - LSTM networks can natively integrate control-flow and data perspectives via concatenated input vectors.
  - Handles repeated activities, categorical attributes, and fewer-example tasks.
---

# Navarin et al. 2017 — Data-Aware LSTM for Remaining Time Prediction

Extends [[concepts/lstm-ppm|LSTM-PPM]] to the **data-aware remaining-time regression** setting — the natural evolution of Polato et al. 2014's classical data-aware model ([[sources/2014-polato-data-aware-remaining-time]]) into the neural era.

## Contribution
- LSTM architecture accepting concatenated activity + event-attribute vectors at each time step.
- Handles repeated activities, categorical attributes, and paths with few training examples.
- Competitive with or superior to tailored baselines on benchmark logs.

## Significance
Together with Tax et al. 2017 ([[sources/2017-tax-lstm-process-prediction]]) and Evermann et al. 2017 ([[sources/2017-evermann-deep-learning-runtime]]), anchors the **2017 neural-PPM trio** — the moment deep learning decisively entered PPM.

## Cited by
- [[sources/2023-riess-temporal-loss-remaining-cycle-time]] — Riess uses the Navarin-style LSTM stack as the baseline architecture when experimenting with temporally weighted L1 losses.
- [[sources/2024-riess-synbps-simulation-framework]] — cited as canonical PPM remaining-time model motivating the need for data-generating-process control.

## Connections
**Concepts:** [[concepts/lstm-ppm]] · [[concepts/remaining-time-prediction]] · [[concepts/trace-encoding]]