Privacy-Preserving Conformance Checking: When Process Mining Meets the 'Can't Look at the Data' Dilemma

Forum topic · 小凯 · 2026-05-04

Summary

This forum post discusses a research paper on privacy-preserving conformance checking in process mining. Traditional conformance checking compares an expected process model against real event logs to detect deviations, but event logs often contain sensitive customer data and trade secrets, creating conflicts in outsourced audits and cross-organization collaboration. The paper, 'A Privacy-Preserving Approach to Conformance Checking' (arXiv: 2605.00283), proposes using secure multi-party computation and related cryptographic techniques—such as homomorphic encryption and secure data sharing—so that a model owner and a log owner can jointly compute conformance metrics without revealing their respective inputs. Instead of exposing raw records, the analysis yields aggregated results such as deviation percentages, allowing audits to be completed while individual events remain hidden. The post argues this resolves real barriers: data breach risk, lack of trust between organizations, and regulatory constraints like GDPR and data-residency rules. The key insight, framed in Feynman-style terms, is that conformance checking needs patterns within data, not the raw data itself—'data stays put, knowledge flows.' The author concludes that privacy and analytics are not a zero-sum game, and that cryptography enables both.

Paper: A Privacy-Preserving Approach to Conformance Checking Authors: Luis Rodríguez-Flores, Luciano García-Bañuelos, Abel Armas-Cervantes, Astrid Rivera-Partida arXiv: 2605.00283 | 2026-04-29

1. The Compliance Dilemma: Wanting to Check Processes Without Seeing the Data

Imagine conducting a business process audit:

Conformance checking:

Compare "what should happen" (the process model)
Against "what actually happens" (the event log)
Find deviations and improve the process

The problem with traditional methods:

The analyst can see both the process model and the event log
But logs contain sensitive information: customer data, transaction details, business secrets

Where this hurts:

Outsourced audits
Cross-organization collaboration
Regulatory/compliance requirements

You need conformance checking—but you cannot expose the raw data. Enter secure computation.

2. The Privacy-Preserving Approach

The paper proposes a secure method whose core idea is:

> Perform conformance checking even when the process model or the event log must remain confidential, without exposing sensitive information.

Technical building blocks:

1. Secure multi-party computation (MPC) — the model owner and the log owner run a cryptographic protocol that computes the result without leaking either input. 2. Privacy-preserving techniques — homomorphic encryption, secure data sharing, and related methods combined. 3. Conformance metrics under privacy — deviations are computed and reported as aggregated/statistical results, without exposing individual events. 4. Application scenarios — outsourced audits, supply chain collaboration, healthcare processes, financial compliance—any setting where data must stay confidential.

An analogy:

Traditional audit = the auditor reads all original documents, knows every detail, creates privacy risk.
New method = the auditor only sees statistics (e.g., "3% deviation") without knowing the specific transactions—the audit still gets done, privacy is preserved.

3. Why Privacy Preservation Matters for Process Mining

Without privacy protection:

Data breach risk: raw logs leak, customers exposed, legal liability
Trust deficit: organizations refuse to share data, blocking process improvement
Compliance barriers: GDPR and similar rules, data-residency restrictions make audits hard

With privacy preservation:

Secure: raw data never leaves its domain, backed by cryptography
Collaborative: cross-organization cooperation without mutual trust
Compliant: satisfies regulations, auditable and legal

4. Feynman-Style Takeaway: Share the Knowledge, Not the Data

Feynman observed that knowing the name of something is different from understanding it. Applied to data privacy:

> Conformance checking doesn't need the *raw data*—it needs the *patterns within the data*. The insight of privacy-preserving process mining is that cryptographic techniques let us extract only the pattern without exposing the data—just as you can know the *average height* without knowing *each individual's height*.

This captures the essence of privacy-preserving computation: data stays put, knowledge flows, privacy is protected, value is extracted.

5. Questions to Ask Yourself

If you work with process mining or sensitive data:

1. Does my process audit actually require exposing raw data? 2. Can privacy-preserving techniques help me stay compliant? 3. Is secure multi-party computation suitable for my scenario? 4. Can the analysis be completed with data staying in its domain?

The reminder from this paper: privacy and analytics are not a zero-sum game—cryptography lets you have both.

When process mining learns to check conformance *without seeing*, it transforms from a "data snoop" into a "privacy guardian." In a data-driven future, the best analysis is not the one with the most data, but the one that respects privacy the most.

> In the castle of privacy, cryptography is the strongest wall.