CoChat

One-Class Anomaly Detection for Industrial Applications: A Comparative Survey and Experimental Study

Open access
Paolini, Davide, Dini, Pierpaolo, Soldaini, Ettore, Saponara, Sergio · 2025 · Computers

The DOI matches its Crossref record; 194 of the 195 references it cites were found in a registry or an index, and 1 are not checked yet.

Abstract

This article aims to evaluate the runtime effectiveness of various one-class classification (OCC) techniques for anomaly detection in an industrial scenario reproduced in a laboratory setting. To address the limitations posed by restricted access to proprietary data, the study explores OCC methods that learn solely from legitimate network traffic, without requiring labeled malicious samples. After analyzing major publicly available datasets, such as KDD Cup 1999 and TON-IoT, as well as the most widely used OCC techniques, a lightweight and modular intrusion detection system (IDS) was developed in Python. The system was tested in real time on an experimental platform based on Raspberry Pi, within a simulated client–server environment using the NFSv4 protocol over TCP/UDP. Several OCC models were compared, including One-Class SVM, Autoencoder, VAE, and Isolation Forest. The results showed strong performance in terms of detection accuracy and low latency, with the best outcomes achieved using the UNSW-NB15 dataset. The article concludes with a discussion of additional strategies to enhance the runtime analysis of these algorithms, offering insights into potential future applications and improvement directions.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.3390/computers14070281

Cites

194 of 195 references verified0 citation errors · 0 not checkable

Statistics

Figures

Figure 1 KDD-99 CUP train and test data distribution.
Figure 1 KDD-99 CUP train and test data distribution.CC-BY · From the PDF
Figure 2 NSL-KDD 2009 train and test data distribution.
Figure 2 NSL-KDD 2009 train and test data distribution.CC-BY · From the PDF
Figure 3 CICIDS 2017 data distribution.
Figure 3 CICIDS 2017 data distribution.CC-BY · From the PDF
Figure 4 UNSW-NB15 2015 train and test data distribution.
Figure 4 UNSW-NB15 2015 train and test data distribution.CC-BY · From the PDF
Figure 5 BoT-IoT 2019 data distribution.
Figure 5 BoT-IoT 2019 data distribution.CC-BY · From the PDF
Figure 6 TON-IoT 2020 data distribution.
Figure 6 TON-IoT 2020 data distribution.CC-BY · From the PDF
Figure 7 Structural flow diagram of the Autoencoder (AE) used for anomaly detection. The input feature vector is encoded into a latent representation, then reconstructed and compared to the original input to compute the reconstruction error. High reconstruction error indicates potential anomalies.
Figure 7 Structural flow diagram of the Autoencoder (AE) used for anomaly detection. The input feature vector is encoded into a latent representation, then reconstructed and compared to the original input to compute the reconstruction error. High reconstruction error indicates potential anomalies.CC-BY · From the PDF
Figure 8 Architectural diagram of the Variational Autoencoder (VAE) process for one-class anomaly detection. The encoder maps input samples to a latent distribution, from which latent vectors are sampled and passed through the decoder to reconstruct the input. During inference, anomalies are identified based on high reconstruction error or low likelihood under the learned distribution of normal data.
Figure 8 Architectural diagram of the Variational Autoencoder (VAE) process for one-class anomaly detection. The encoder maps input samples to a latent distribution, from which latent vectors are sampled and passed through the decoder to reconstruct the input. During inference, anomalies are identified based on high reconstruction error or low likelihood under the learned distribution of normal data.CC-BY · From the PDF
Figure 9 Architectural diagram of the Isolation Forest (iForest) process for one-class anomaly detection. The algorithm constructs multiple binary trees by recursively partitioning the input data using randomly selected features and split values. Normal instances, which are densely clustered, require more partitions to isolate, resulting in longer average path lengths. In contrast, anomalies are isolated earlier due to their sparsity, yielding shorter path lengths.
Figure 9 Architectural diagram of the Isolation Forest (iForest) process for one-class anomaly detection. The algorithm constructs multiple binary trees by recursively partitioning the input data using randomly selected features and split values. Normal instances, which are densely clustered, require more partitions to isolate, resulting in longer average path lengths. In contrast, anomalies are isolated earlier due to their sparsity, yielding shorter path lengths.CC-BY · From the PDF
Figure 10 Architectural diagram of the One-Class Support Vector Machine (OCSVM) process for anomaly detection. The model learns a decision boundary that best encloses the distribution of normal data in a high-dimensional feature space using a kernel function. During inference, samples that fall outside the learned boundary are considered anomalies.
Figure 10 Architectural diagram of the One-Class Support Vector Machine (OCSVM) process for anomaly detection. The model learns a decision boundary that best encloses the distribution of normal data in a high-dimensional feature space using a kernel function. During inference, samples that fall outside the learned boundary are considered anomalies.CC-BY · From the PDF
Figure 11 Matrix correlation of NSL-KDD 2009 dataset.
Figure 11 Matrix correlation of NSL-KDD 2009 dataset.CC-BY · From the PDF
Figure 12 Features reduction phase of NSL-KDD 2009 dataset.
Figure 12 Features reduction phase of NSL-KDD 2009 dataset.CC-BY · From the PDF
Figure 13 Correlation matrix of the UNSW-NB15 dataset after preprocessing. Color intensity indicates the strength and direction of linear correlation between pairs of features. Highly correlated features are candidates for removal or PCA transformation.
Figure 13 Correlation matrix of the UNSW-NB15 dataset after preprocessing. Color intensity indicates the strength and direction of linear correlation between pairs of features. Highly correlated features are candidates for removal or PCA transformation.CC-BY · From the PDF
Figure 14 Features reduction phase of UNSW-NB15 2009 dataset.
Figure 14 Features reduction phase of UNSW-NB15 2009 dataset.CC-BY · From the PDF
Figure 15 Matrix correlation of CICIDS 2017 dataset.
Figure 15 Matrix correlation of CICIDS 2017 dataset.CC-BY · From the PDF
Figure 16 Features reduction phase of CICIDS 2017 dataset.
Figure 16 Features reduction phase of CICIDS 2017 dataset.CC-BY · From the PDF
Figure 17 Matrix correlation of TON-IoT dataset.
Figure 17 Matrix correlation of TON-IoT dataset.CC-BY · From the PDF
Figure 18 Features reduction phase of TON-IoT dataset.
Figure 18 Features reduction phase of TON-IoT dataset.CC-BY · From the PDF
Figure 19 Architecture of the experimental setup: NFSv4 server with runtime IDS and three clients.
Figure 19 Architecture of the experimental setup: NFSv4 server with runtime IDS and three clients.CC-BY · From the PDF
Image 20
CC-BY · From the PDF
Figure 21 The generic GUI interface displayed before initiating the IDS is shown on the right, while on the left are the system behaviors in the absence of anomalies (A) and in the presence of anomalies (B).
Figure 21 The generic GUI interface displayed before initiating the IDS is shown on the right, while on the left are the system behaviors in the absence of anomalies (A) and in the presence of anomalies (B).CC-BY · From the PDF
Figure 22 Experimental edge network setup with NFSv4 and ML tested.
Figure 22 Experimental edge network setup with NFSv4 and ML tested.CC-BY · From the PDF
Figure 23 Boxplot of runtime inference time distribution per packet for each OCC algorithm tested.
Figure 23 Boxplot of runtime inference time distribution per packet for each OCC algorithm tested.CC-BY · From the PDF

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Research works printed but not linked: 5

  1. Khan, 2024, The Future of Cybersecurity: Leveraging Artificial Intelligence to Combat Evolving Threats and Enhance Digital Defense Strategies, J. Comput. Anal. Appl., 33, 776
  2. Hettich, S., and Bay, S. (1999). The UCI KDD Archive, University of California, Department of Information and Computer Science. Available online: http://kdd.ics.uci.edu/.
  3. Aleesa, 2021, Deep-intrusion detection system with enhanced UNSW-NB15 dataset based on deep learning techniques, J. Eng. Sci. Technol., 16, 711
  4. Böhm, V., and Seljak, U. (2020). Probabilistic autoencoder. arXiv.
  5. Pandey, 2024, A KNN-Based Intrusion Detection System for Enhanced Anomaly Detection in Industrial IoT Networks, Int. J. Innov. Res. Technol. Sci., 12, 1

Cited by

See all 21

Metadata

Authors
Paolini, D., Dini, P., Soldaini, E., & Saponara, S.
Affiliations
  • University of Pisa (IT)
  • Vrije Universiteit Brussel (BE)
Corresponding author
Dini, Pierpaolo
Venue
Computers
Publisher
MDPI AG
Published
2025-07-16
Volume
14
Issue
7
Pages
281
ISSN
2073-431X (ISSN-L)
OpenAlex
W4412476711
Cited by
21
References
188
Citations by year
2026: 12 · 2025: 9
FWCI
8.88
Type
journal article
Language
en
Open access
Locations
  • Computers (publishedVersion, cc-by)
  • CINECA IRIS Institutial research information system (University of Pisa) (submittedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
  • Project PNRR CN 1 “Piano Nazionale di Ripresa e Resilineza del Centro Nazionale 1: Simulation, Calculation and Analysis of High-Performance Data” (I53C22000690001, 800928, 101036168)
  • European High-Performance Computing Joint Undertaking (JU) (I53C22000690001, 800928, 101036168)
  • Italian Ministry of Education and Research (MIUR) (I53C22000690001, 800928, 101036168)
Topics
  • Physical Sciences › Computer Science › Computer Networks and Communications › Network Security and Intrusion Detection
  • Physical Sciences › Computer Science › Artificial Intelligence › Anomaly Detection Techniques and Applications
  • Physical Sciences › Computer Science › Artificial Intelligence › Artificial Immune Systems Applications
Keywords
one-class classification, anomaly detection, intrusion detection, industrial networks, network traffic classification, one-class support vector machine, autoencoder, variational autoencoder, Isolation Forest, KDD Cup 1999 dataset, ToN-IoT, UNSW-NB15, NFSv4, real-time detection, Raspberry Pi

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.