---
title: "Exploiting Event Log Event Attributes in RNN Based Prediction"
type: source
tags: [ppm, rnn, data-aware, clustering, scalability]
authors: [Hinkka Markku; Lehto Teemu; Heljanko Keijo]
year: 2019
venue: "arXiv:1904.06895v2"
kind: paper
raw_path: "raw/Predictive process monitoring/Exploiting Event Log Event Attributes in RNN.pdf"
created: 2026-04-13
updated: 2026-04-13
key_claims:
  - Prior RNN-PPM work underused event attributes; the challenge is scalability with many attributes/values.
  - Cluster events by their attribute values and feed cluster labels (rather than raw attributes) to the RNN.
  - Combining clustering with raw-attribute RNN input can further improve accuracy at the cost of training/prediction time.
---

# Hinkka, Lehto, Heljanko 2019 — Attribute Clustering for RNN-PPM

Addresses the **scalability** problem of data-aware [[concepts/lstm-ppm|RNN-PPM]]: when event attributes have many values, one-hot encoding explodes the input dimensionality.

## Contribution
- Pre-clusters events by their attribute-value signature.
- Feeds cluster labels (bounded cardinality) to the RNN instead of raw attributes.
- Trade-off knob between accuracy and training cost.

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