---
title: "An Eye into the Future: Leveraging A-priori Knowledge in Predictive Business Process Monitoring"
type: source
tags: [ppm, lstm, a-priori-knowledge, suffix-prediction]
authors: [Di Francescomarino Chiara; Ghidini Chiara; Maggi Fabrizio Maria; Petrucci Giulio; Yeshchenko Anton]
year: 2017
venue: "BPM 2017, LNCS 10445, Springer: 252–268"
kind: paper
raw_path: "raw/Predictive process monitoring/An Eye into the Future Leveraging A-priori difrancescomarino2017.pdf"
created: 2026-04-13
updated: 2026-04-13
key_claims:
  - A-priori case-specific knowledge about the future can be fed to an LSTM predictor at inference time (without retraining).
  - Two novel techniques — NOCYCLE (cycle-avoidance) and A-PRIORI (future-constraint-aware) — improve prediction accuracy up to 50% over plain LSTM baselines.
  - Evaluated on six real-life logs.
---

# Di Francescomarino et al. 2017 — A-priori Knowledge in PPM

Extends [[concepts/lstm-ppm|LSTM-PPM]] to incorporate **a-priori knowledge** about how a case *will* unfold — e.g., known future constraints like "surgery room unavailable at time T".

## Contribution
- **NOCYCLE** — avoids the local-minimum failure mode of LSTM suffix prediction where the model loops indefinitely.
- **A-PRIORI** — takes future constraints as runtime input, steering the LSTM's suffix prediction to remain compatible.
- Crucially: a-priori knowledge is handled *at inference*, not via retraining — so new constraints can be supplied on-the-fly.

## Significance
Demonstrates that neural PPM need not be a closed black box; external knowledge (typically LTL-like constraints) can meaningfully improve predictions. Aligns with the APM Manifesto's [[concepts/framed-autonomy|framed autonomy]] intuition — predictors constrained by normative frames.

## Connections
**Concepts:** [[concepts/predictive-process-monitoring]] · [[concepts/next-activity-prediction]] · [[concepts/lstm-ppm]]
**Authors:** [[entities/chiara-di-francescomarino]] · [[entities/chiara-ghidini]] · [[entities/fabrizio-maggi]]
**Related sources:** [[sources/2017-tax-lstm-process-prediction]]