MI$^2$DAS: A Multi-Layer Intrusion Detection Framework with Incremental Learning for Securing Industrial IoT Networks

Wei Lian; Alejandro Guerra-Manzanares

cs.CRFeb 27, 2026

MI $^2$ DAS: A Multi-Layer Intrusion Detection Framework with Incremental Learning for Securing Industrial IoT Networks

Authors: Wei Lian, Alejandro Guerra-Manzanares

Organizations: School of Computer Science, University of Nottingham, Ningbo, China

Abstract

The rapid expansion of Industrial IoT (IIoT) systems has amplified security challenges, as heterogeneous devices and dynamic traffic patterns increase exposure to sophisticated and previously unseen cyberattacks. Traditional intrusion detection systems often struggle in such environments due to their reliance on extensive labeled data and limited ability to detect new threats. To address these challenges, we propose MI $^2$ DAS, a multi-layer intrusion detection framework that integrates anomaly-based hierarchical traffic pooling, open-set recognition to distinguish between known and unknown attacks and incremental learning for adapting to novel attack types with minimal labeling. Experiments conducted on the Edge-IIoTset dataset demonstrate strong performance across all layers. In the first layer, GMM achieves superior normal-attack discrimination (accuracy = 0.953, TPR = 1.000). In open-set recognition, GMM attains a recall of 0.813 for known attacks, while LOF achieves 0.882 recall for unknown attacks. For fine-grained classification of known attacks, Random Forest achieves a macro-F1 of 0.941. Finally, the incremental learning module maintains robust performance when incorporation novel attack classes, achieving a macro-F1 of 0.8995. These results showcase MI $^2$ DAS as an effective, scalable and adaptive framework for enhancing IIoT security against evolving threats.

Figures & tables

Figure 1: Architecture of the Multi-Layer IIoT Intrusion Detection Adaptive System (MI 2 DAS).

Class	Total	Training	Test
Normal	24301	19281	4820
Backdoor	10195	7892	1973
DDoS_HTTP	10561	8396	2099
DDoS_ICMP	14090	10477	2619
DDoS_TCP	10247	8198	2049
DDoS_UDP	14498	11598	2900

Table 1: Dataset class distribution ( Ferrag et al.,, 2022 )

Setting	Parameter	Acc.	TPR	FPR	Pr.
Novelty	nc=2, th_per=5	0.953	1.000	0.095	0.914
	nc=4, th_per=5	0.950	1.000	0.099	0.910
	nc=3, th_per=5	0.950	1.000	0.100	0.909
Outlier	nc=2, th_per=5	0.944	1.000	0.112	0.899
	nc=4, th_per=5	0.810	1.000	0.381	0.724
	nc=3, th_per=5	0.804	1.000	0.392	0.718

Table 2: Top-3 GMM performance per setting. The parameter nc denotes the number of mixture components, while th_per indicates the threshold percentile used for anomaly classification. The best results are highlighted in bold.

Figure 2: Recall performance for GMM and LOF methods for GMM and LOF methods

Model	Macro-F1	Weighted-F1	Macro-Accuracy	Micro-Accuracy
RF	0.941 $\pm$ 0.054	0.944 $\pm$ 0.056	0.950 $\pm$ 0.048	0.949 $\pm$ 0.050
KNN	0.938 $\pm$ 0.033	0.938 $\pm$ 0.036	0.937 $\pm$ 0.034	0.938 $\pm$ 0.040
XGBoost	0.906 $\pm$ 0.079	0.910 $\pm$ 0.083	0.924 $\pm$ 0.068	0.920 $\pm$ 0.072
LightGBM	0.905 $\pm$ 0.079	0.908 $\pm$ 0.083	0.922 $\pm$ 0.068	0.918 $\pm$ 0.073
SVM	0.866 $\pm$ 0.064	0.883 $\pm$ 0.063	0.882 $\pm$ 0.059	0.885 $\pm$ 0.060
LR	0.862 $\pm$ 0.063	0.881 $\pm$ 0.061	0.887 $\pm$ 0.057	0.882 $\pm$ 0.059

Table 3: Aggregated results across all combinations and evaluated algorithms. The best results are highlighted in bold.

Figure 3: Macro F1 performance values for different classification models and Known-Unknown combinations

Method	Macro F1	Balanced Acc	Acc
Self-training	0.8859 $\pm$ 0.0182	0.8954 $\pm$ 0.0183	0.8995 $\pm$ 0.0140
Label Spreading	0.8349 $\pm$ 0.0337	0.8561 $\pm$ 0.0258	0.8590 $\pm$ 0.0354
Label Propagation	0.8404 $\pm$ 0.0210	0.8632 $\pm$ 0.0227	0.8630 $\pm$ 0.0186
Active Learning	0.8296 $\pm$ 0.0353	0.8561 $\pm$ 0.0215	0.8548 $\pm$ 0.0254

Table 4: Average performance of One-Step Iteration for

N=4

Known Attacks (5 random runs). The best results are highlighted in bold.

Method	Macro F1	Balanced Acc	Acc
Self-training	0.8648 $\pm$ 0.0090	0.8908 $\pm$ 0.0127	0.8834 $\pm$ 0.0100
Label Spreading	0.7788 $\pm$ 0.0299	0.8188 $\pm$ 0.0274	0.8065 $\pm$ 0.0334
Label Propagation	0.8007 $\pm$ 0.0373	0.8350 $\pm$ 0.0326	0.8241 $\pm$ 0.0379
Active Learning	0.8113 $\pm$ 0.0240	0.8553 $\pm$ 0.0144	0.8504 $\pm$ 0.0174

Table 5: Average performance of One-Step Iteration for

N=7

Known Attacks (5 random runs). The best results are highlighted in bold.

Method	Macro F1	Balanced Acc	Acc
Self-training	0.8617 $\pm$ 0.0102	0.8835 $\pm$ 0.0119	0.8822 $\pm$ 0.0076
Label Spreading	0.7753 $\pm$ 0.0554	0.8044 $\pm$ 0.0424	0.8042 $\pm$ 0.0472
Label Propagation	0.8019 $\pm$ 0.0426	0.8270 $\pm$ 0.0404	0.8280 $\pm$ 0.0379
Active Learning	0.8303 $\pm$ 0.0185	0.8664 $\pm$ 0.0221	0.8655 $\pm$ 0.0224

Table 6: Average performance of One-Step Iteration for

N=10

Known Attacks (5 random runs). The best results are highlighted in bold.

Method	Macro F1	Balanced Ac c	Acc
Self-training	0.8970 $\pm$ 0.0089	0.9023 $\pm$ 0.0061	0.9053 $\pm$ 0.0066
Label Spreading	0.8508 $\pm$ 0.0382	0.8752 $\pm$ 0.0341	0.8692 $\pm$ 0.0371
Label Propagation	0.8590 $\pm$ 0.0219	0.8836 $\pm$ 0.0206	0.8765 $\pm$ 0.0222
Active Learning	0.8128 $\pm$ 0.0169	0.8377 $\pm$ 0.0113	0.8367 $\pm$ 0.0119

Table 7: Average performance of One-Step Iteration for

N=13

Known Attacks (5 random runs). The best results are highlighted in bold.

Step	Strategy	Macro-F1	Balanced Acc	Accuracy
First Step (4+10)	Both	0.9133	0.9230	0.9215
Second Step (7+7)	Seed-based	0.8873	0.8965	0.8965
Second Step (7+7)	Augmentation	0.9085	0.9150	0.9158
Third Step (10+4)	Seed-based	0.8741	0.8894	0.8900
Third Step (10+4)	Augmentation	0.8860	0.9021	0.9004
Final Step (13+1)	Seed-based	0.8637	0.8847	0.8830

Table 8: Performance results of multi-step iteration for different strategies. The best results per step are highlighted in bold.

Explore similar work

Toward a Multi-Layer ML-Based Security Framework for Industrial IoT
Mar 25, 2026Aymen Bouferroum, Valeria Loscri, Abderrahim BenslimaneInternet of Things IoT Intrusion Detection
LIA · Inria Lille-Nord Europe, Lille, France · LIA/CERI University of Avignon, Avignon, France
CLAD: A Clustered Label-Agnostic Federated Learning Framework for Joint Anomaly Detection and Attack Classification
May 7, 2026Iason Ofeidis, Nikos Papadis, Randeep Bhatia +2IoT Intrusion Detection Network Intrusion Detection
Yale University, New Haven, CT, USA · Nokia Bell Labs, Murray Hill, NJ, USA
SE-Enhanced ViT and BiLSTM-Based Intrusion Detection for Secure IIoT and IoMT Environments
Apr 6, 2026Afrah Gueriani, Hamza Kheddar, Ahmed Cherif Mazari +2IoT Intrusion Detection Recurrent Transformers
LSEA Lab., Faculty of Technology, University of MEDEA, Medea 26000, Algeria · LSEA Lab, Faculty of Science, University of MEDEA, Medea 26000, Algeria · Department of Computer Engineering, Gazi University, Ankara, Turkey +1

Mar 25, 2026J/K move · Enter open · S save

Toward a Multi-Layer ML-Based Security Framework for Industrial IoT

Aymen Bouferroum, Valeria Loscri, Abderrahim Benslimane

LIA · Inria Lille-Nord Europe, Lille, France · LIA/CERI University of Avignon, Avignon, France

Internet of Things IoT Intrusion Detection

Open paper AskPDF arXiv

The Industrial Internet of Things (IIoT) introduces significant security challenges as resource-constrained devices become increasingly integrated into critical industrial processes. Existing security approaches typically address threats at a single network layer, often relying on expensive hardware and remaining confined to simulation environments. In this paper, we present the research framework and contributions of our doctoral thesis, which aims to develop a lightweight, Machine Learning (ML)-based security framework for IIoT environments. We first describe our adoption of the Tm-IIoT trust model and the Hybrid IIoT (H-IIoT) architecture as foundational baselines, then introduce the Trust Convergence Acceleration (TCA) approach, our primary contribution that integrates ML to predict and mitigate the impact of degraded network conditions on trust convergence, achieving up to a 28.6% reduction in convergence time while maintaining robustness against adversarial behaviors. We then propose a real-world deployment architecture based on affordable, open-source hardware, designed to implement and extend the security framework. Finally, we outline our ongoing research toward multi-layer attack detection, including physical-layer threat identification and considerations for robustness against adversarial ML attacks.

Abstract

Explore similar work

Toward a Multi-Layer ML-Based Security Framework for Industrial IoT

CLAD: A Clustered Label-Agnostic Federated Learning Framework for Joint Anomaly Detection and Attack Classification

SE-Enhanced ViT and BiLSTM-Based Intrusion Detection for Secure IIoT and IoMT Environments