Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Towards Transparent and Efficient Anomaly Detection in Industrial Processes through ExIFFI

Davide Frizzo, Francesco Borsatti, Alessio Arcudi, Antonio De Moliner, Roberto Oboe, Gian Antonio Susto

TL;DR

This work tackles the need for transparent anomaly detection in industrial processes by presenting ExIFFI, a fast, model-specific explanation framework for Extended Isolation Forest (EIF) and its extension EIFp. ExIFFI provides both Global Feature Importance ($GFI$) and Local Feature Importance ($LFI$) by leveraging EIF’s tree structure, with per-node and per-tree contributions aggregated across the forest and normalized to correct biases. On three industrial datasets, ExIFFI delivers high anomaly detection performance (e.g., averaging over 90% AP) while outperforming post-hoc XAI methods in feature-selection proxy tasks, and it achieves substantial speed advantages over alternatives like KernelSHAP. The approach supports real-time, human-centered decision-making in Industry 5.0 by delivering efficient, interpretable explanations that align with domain insight, though it remains tied to EIF-based models and may face challenges with clustered anomalies.

Abstract

Anomaly Detection (AD) is crucial in industrial settings to streamline operations by detecting underlying issues. Conventional methods merely label observations as normal or anomalous, lacking crucial insights. In Industry 5.0, interpretable outcomes become desirable to enable users to understand the rational under model decisions. This paper presents the first industrial application of ExIFFI, a recent approach for fast, efficient explanations for the Extended Isolation Forest (EIF) AD method. ExIFFI is tested on three industrial datasets, demonstrating superior explanation effectiveness, computational efficiency and improved raw anomaly detection performances. ExIFFI reaches over then 90\% of average precision on all the benchmarks considered in the study and overperforms state-of-the-art Explainable Artificial Intelligence (XAI) approaches in terms of the feature selection proxy task metric which was specifically introduced to quantitatively evaluate model explanations.

Towards Transparent and Efficient Anomaly Detection in Industrial Processes through ExIFFI

TL;DR

This work tackles the need for transparent anomaly detection in industrial processes by presenting ExIFFI, a fast, model-specific explanation framework for Extended Isolation Forest (EIF) and its extension EIFp. ExIFFI provides both Global Feature Importance () and Local Feature Importance () by leveraging EIF’s tree structure, with per-node and per-tree contributions aggregated across the forest and normalized to correct biases. On three industrial datasets, ExIFFI delivers high anomaly detection performance (e.g., averaging over 90% AP) while outperforming post-hoc XAI methods in feature-selection proxy tasks, and it achieves substantial speed advantages over alternatives like KernelSHAP. The approach supports real-time, human-centered decision-making in Industry 5.0 by delivering efficient, interpretable explanations that align with domain insight, though it remains tied to EIF-based models and may face challenges with clustered anomalies.

Abstract

Anomaly Detection (AD) is crucial in industrial settings to streamline operations by detecting underlying issues. Conventional methods merely label observations as normal or anomalous, lacking crucial insights. In Industry 5.0, interpretable outcomes become desirable to enable users to understand the rational under model decisions. This paper presents the first industrial application of ExIFFI, a recent approach for fast, efficient explanations for the Extended Isolation Forest (EIF) AD method. ExIFFI is tested on three industrial datasets, demonstrating superior explanation effectiveness, computational efficiency and improved raw anomaly detection performances. ExIFFI reaches over then 90\% of average precision on all the benchmarks considered in the study and overperforms state-of-the-art Explainable Artificial Intelligence (XAI) approaches in terms of the feature selection proxy task metric which was specifically introduced to quantitatively evaluate model explanations.
Paper Structure (22 sections, 6 equations, 6 figures, 6 tables)

This paper contains 22 sections, 6 equations, 6 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Proposed Approach
  4. Experimental Results
  5. Industrial IoT datasets
  6. CoffeeData Dataset
  7. Experimental Setup
  8. AD Performance Comparison
  9. Case Study I: TEP Dataset
  10. Global Feature Importance
  11. Feature Selection Proxy Task
  12. Case Study II: PIADE Dataset
  13. Global Feature Importance
  14. Case Study III: CoffeeData Dataset
  15. Global Feature Importance
  16. ...and 7 more sections

Figures (6)

  • Figure 1: This plot represents the top 8 features of the GFI ranking returned by ExIFFI on TEP.
  • Figure 2: Feature selection results for inverse (red), direct (blue), and random (green) approaches using DIFFI, AcME-AD, KernelSHAP, and ExIFFI XAI: (a) TEP, (b) CoffeeData.
  • Figure 3: This plot represents the top 8 features of the GFI ranking returned by ExIFFI on PIADE.
  • Figure 4: The trend of the GFI scores increases in the time interval where there is the most significant difference between the time series of normal and anomalous capsules
  • Figure 5: The average precision metric increases as the number of trees used to fit the AD model increases
  • ...and 1 more figures