---
title: "Business Process Remaining Time Prediction Using Explainable Reachability Graph from Gated RNNs"
type: source
tags: [ppm, gated-rnn, explainable, petri-net, remaining-time]
authors: [Cao Rui; Zeng Qingtian; Ni Weijian; Duan Hua; Liu Cong; Lu Faming; Zhao Ziqi]
year: 2023
venue: "Applied Intelligence 53: 13178–13191, Springer"
kind: paper
raw_path: "raw/Predictive process monitoring/gated rnns.pdf"
created: 2026-04-13
updated: 2026-04-13
key_claims:
  - Gated RNNs (LSTM/GRU) for remaining-time prediction are black boxes; prefix-bucketing methods train one predictor per bucket (less accurate).
  - A single gated RNN on the whole log can be explained by mapping its hidden states to states of an explicit reachability graph of a discovered Petri net.
  - The mapping yields an interpretable remaining-time predictor without sacrificing accuracy.
  - Evaluated on six real-life logs.
---

# Cao et al. 2023 — Explainable Gated RNN via Reachability Graph

Addresses the **interpretability** gap of gated-RNN [[concepts/lstm-ppm|PPM]]. Proposes:
1. Train a gated RNN on the whole log for [[concepts/remaining-time-prediction|remaining-time prediction]].
2. Discover a Petri net from the log; compute its reachability graph.
3. Map the RNN's hidden states to reachability-graph states, producing explanations of what process state drives each prediction.

## Position
Part of the growing **explainable PPM** strand (cf. [[sources/2017-verenich-white-box-flow-analysis]] for an earlier white-box approach). Aligns with [[concepts/explainability-apm|APM explainability]] challenges.

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