Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Explainable AI for Multi-Class Machine Failure Prediction: A Focus on Calibrated Explainability and Model Uncertainty
Jönköping University, School of Engineering, JTH, Department of Computing, Jönköping AI Lab (JAIL).
2025 (English)Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
Abstract [en]

Predictive maintenance has become essential in modern manufacturing, aiming to prevent costlyand potentially hazardous equipment failures. Traditional maintenance strategies are increasingly beingreplaced by AI-driven predictive models. However, despite their accuracy, many of these modelsoperate as black boxes, offering limited insight into their decision-making processes and thereby undermininguser trust. This study proposes a novel framework that combines Explainable AI (XAI) andUncertainty Quantification (UQ) to enhance both the interpretability and trustworthiness of multi-classfailure predictions. We extend the Venn–Abers-based Calibrated Explanations (CE) method to improveits support for multi-class classification. The framework is evaluated using four industrial datasets andbenchmarked against established models and explanation techniques, including SHapley Additive ex-Planations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME). A Gated RecurrentUnit (GRU) model was selected for its effectiveness in handling temporal data and consistently outperformedtraditional machine learning models across several classification evaluation metrics. Surprisingly,calibration techniques such as Platt Scaling, Isotonic Regression, and Venn–Abers did notreliably calibrate the GRU’s probabilistic outputs, suggesting that deep learning models may exhibitinherent calibration in certain contexts. The extended CE method provided the most robust and stableexplanations, outperforming SHAP and LIME in variance-based evaluation metrics. A within-subjectsurvey involving 22 participants with engineering backgrounds showed that explanations combiningprediction probabilities, feature importance, and uncertainty led to higher trust ratings and more informeddecision-making. Participants preferred CE-based explanations and exhibited greater confidenceand contextual awareness in their decisions, particularly when faced with trade-offs betweenproduction costs and failure risks.

Place, publisher, year, edition, pages
2025. , p. 42
National Category
Computer Sciences
Identifiers
URN: urn:nbn:se:hj:diva-68422OAI: oai:DiVA.org:hj-68422DiVA, id: diva2:1968367
External cooperation
Axion Solutions
Subject / course
JTH, Computer Engineering
Supervisors
Examiners
Available from: 2025-06-13 Created: 2025-06-12 Last updated: 2025-10-13Bibliographically approved

Open Access in DiVA

fulltext(615 kB)249 downloads
File information
File name FULLTEXT01.pdfFile size 615 kBChecksum SHA-512
56e9010dadaf74075ce495ab5bd100926264ad3350d90923c933f01f8e6d42c375f39e89811de0abaea00c6bff18f82bdce7b1ef3fae6cd6f5f1d2b59de0944b
Type fulltextMimetype application/pdf

Search in DiVA

By author/editor
Hanna, Ayham
By organisation
Jönköping AI Lab (JAIL)
Computer Sciences

Search outside of DiVA

GoogleGoogle Scholar
Total: 253 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

urn-nbn

Altmetric score

urn-nbn
Total: 1496 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf