---
title: "Deep Learning Process Prediction with Discrete and Continuous Data Features"
type: source
tags: [ppm, lstm, data-aware, next-activity]
authors: [Schönig Stefan; Jasinski Richard; Ackermann Lars; Jablonski Stefan]
year: 2018
venue: "ENASE/ICSOFT 2018 (SCITEPRESS)"
kind: paper
raw_path: "raw/Predictive process monitoring/Deep learning process prediction with discrete and continuous data features.pdf"
created: 2026-04-13
updated: 2026-04-13
key_claims:
  - Prior LSTM-PPM work focused on activity labels + resources; discrete and continuous data attributes were neglected.
  - Incorporating event-level data attributes significantly improves next-activity prediction accuracy.
  - Validated on a recent real-life event log.
---

# Schönig et al. 2018 — Discrete + Continuous Data Features for LSTM-PPM

Extends LSTM-PPM with **full event-attribute support** — both discrete (categorical) and continuous (numerical). Argues that prior neural PPM work underused event-data features.

## Contribution
- Systematic extension of [[concepts/lstm-ppm|LSTM-PPM]] for multi-perspective data.
- Encoding strategy for mixed discrete/continuous attributes alongside activity labels and resources.
- Empirical gains on a real-life log.

## Position
Part of the 2018 wave of papers that generalised Tax et al. 2017 ([[sources/2017-tax-lstm-process-prediction]]) with richer data-perspective features — alongside [[sources/2017-navarin-lstm-data-aware-remaining-time]] and Camargo et al.

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