Melola Discovery

← Melola Discovery

Akses terbuka Jurnal SINTA 1

An Explainable Continual-Learning Transformer for Real-Time Anomaly Detection in 5G/6G Network Slices

Journal of Robotics and Control (JRC) · Vol. 7 No. 4 (2026) · hlm. 4067–4083

DOI
10.18196/jrc.v7i4.28510
ISSN
2715-5072, 2715-5056
Terbit
2026-07-01
Penerbit
Universitas Muhammadiyah Yogyakarta
Bahasa
eng
Lisensi
creativecommons.org/licenses/by-sa/4.0

Abstrak

The evolution of mobile networks from 5G to 6G enables ultra-reliable, low-latency, and massively connected services through network slicing, yet it simultaneously exposes a broad and continuously shifting attack surface. Static and rule-based anomaly detectors generalize poorly across heterogeneous slices and cannot adapt to previously unseen attacks in real time, while conventional deep models forget earlier knowledge when retrained on new traffic. The research contribution is a unified, explainable, continual-learning Transformer that jointly delivers high detection accuracy, lifelong adaptability, and operator-facing interpretability for 5G/6G slicing. Methodologically, a multi-head self-attention encoder models short- and long-range dependencies in high-dimensional slice telemetry; Elastic Weight Consolidation (EWC) and an experience-replay buffer mitigate catastrophic forgetting across a defined sequence of attack-introduction tasks; and an explainability layer combines attention heatmaps with SHAP attributions. Three complementary variants (Transformer–LSTM, Graph Attention, and Autoencoder–Transformer ensemble) address diverse anomaly types. On 5G-core and Open-RAN datasets, the framework attains 98.87% accuracy, 98.90% precision, 98.70% recall, and 1.8 ms/sample latency, outperforming LSTM, CNN, and Random Forest baselines (McNemar p < 0.001). Continual-learning evaluation across five sequential tasks yields a low forgetting measure (FM = 0.013) versus 0.212 for naive fine-tuning, and ablation studies isolate the contribution of EWC, replay, and sequence length. Per-slice and per-attack analyses reveal where residual errors concentrate. These results show that integrating Transformer encoding, continual learning, and interpretability produces an accurate, adaptable, and transparent detector suited to live 5G/6G deployment.

Kata kunci

Jaringan kolaborasi para penulis

32 penulis · 104 tautan kepenulisan bersama · 31 lintas organisasi

Arahkan kursor atau ketuk sebuah nama untuk menyorot rekan penulisnya. Klik dua kali untuk membuka halaman penulis.

  • Imam Ja’afar Al-Sadiq University
  • University of Baghdad
  • Ministry of Higher Education and Scientific Research
  • Sohar University
  • Al-Ahliyya Amman University
  • Thamar University
  • National University of Science and Technology
  • Organisasi lain
  • Afiliasi tidak diketahui
  • Organisasi berbeda
  • Organisasi sama
  • Afiliasi tidak diketahui
  • Garis lebih tebal = lebih banyak artikel bersama · lingkaran lebih besar = lebih banyak artikel
Tampilkan sebagai tabel
Rekan penulisOrganisasiArtikel bersama
Faris Hassan TahaUniversity of Mosul2
Abadal-Salam T. HussainUniversiti Malaysia Perlis2
Taha A. TahaNorthern Technical University2
Ahmed Lateef KhalafImam Ja’afar Al-Sadiq University1
Nuha Sami MohsinUniversity of Baghdad1
Saadya Fahad JabbarUniversity of Baghdad1
Mosleh M. AbualhajAl-Ahliyya Amman University1
Alhamza MuntherAfiliasi tidak diketahui1
Alabass AlalousiThamar University1
Haider J. AbdUniversity of Babylon1
Feras Hasson AlattarNational University of Science and Technology1
Tariq Adnan FadilDijlah University College1
Abadal-Salam T. HussainAfiliasi tidak diketahui1
Taha A. TahaAfiliasi tidak diketahui1
Abdulghafor Mohammed HashimAfiliasi tidak diketahui1
Mohammed M AldabbaghAfiliasi tidak diketahui1
Yasin BektaşAfiliasi tidak diketahui1
Basim Ghalib MejbelAfiliasi tidak diketahui1
B. A. AhmadAfiliasi tidak diketahui1
B. S. VeenaAfiliasi tidak diketahui1
Hassan Muwafaq GheniAl-Mustaqbal University1
Zinah S. JabbarImam Ja’afar Al-Sadiq University1
Sura Abdulmunem Mohammed Al-JubooriMinistry of Higher Education and Scientific Research1
Firas HazzaaMinistry of Higher Education and Scientific Research1

Afiliasi berasal dari OpenAlex; metadata OAI-PMH jurnal tidak memuatnya. Penulis tanpa ID OpenAlex dicocokkan berdasarkan nama, sehingga dua orang bisa tergabung atau satu orang bisa terpisah.

Sumber metadata: OAI-PMH (metadata jurnal) https://journal.umy.ac.id/index.php/jrc/oai, pertama di-harvest 9 Nov 2026, terakhir terlihat 16 Nov 2026.
Metadata asli (oai_dc)
<oai_dc:dc xmlns="http://www.openarchives.org/OAI/2.0/" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:oai_dc="http://www.openarchives.org/OAI/2.0/oai_dc/" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:schemaLocation="http://www.openarchives.org/OAI/2.0/oai_dc/  http://www.openarchives.org/OAI/2.0/oai_dc.xsd">
	<dc:title xml:lang="en-US">An Explainable Continual-Learning Transformer for Real-Time Anomaly Detection in 5G/6G Network Slices</dc:title>
	<dc:creator>Zainab J. Taher</dc:creator>
	<dc:creator>Maha I. Raheem</dc:creator>
	<dc:creator>Sinan Q. Salih</dc:creator>
	<dc:creator>Poh Soon JosepNg</dc:creator>
	<dc:creator>Taha Almulaisi</dc:creator>
	<dc:creator>Felix Pasila</dc:creator>
	<dc:creator>Koo Yuen Phan</dc:creator>
	<dc:creator>Hilal A. Fadhil</dc:creator>
	<dc:subject xml:lang="en-US">Network Slicing Security</dc:subject>
	<dc:subject xml:lang="en-US">Artificial Intelligence (AI)</dc:subject>
	<dc:subject xml:lang="en-US">Lifelong Learning for Networks</dc:subject>
	<dc:subject xml:lang="en-US">Transformer Anomaly Detection</dc:subject>
	<dc:subject xml:lang="en-US">Explainable AI</dc:subject>
	<dc:subject xml:lang="en-US">6G Adaptive Security</dc:subject>
	<dc:description xml:lang="en-US">The evolution of mobile networks from 5G to 6G enables ultra-reliable, low-latency, and massively connected services through network slicing, yet it simultaneously exposes a broad and continuously shifting attack surface. Static and rule-based anomaly detectors generalize poorly across heterogeneous slices and cannot adapt to previously unseen attacks in real time, while conventional deep models forget earlier knowledge when retrained on new traffic. The research contribution is a unified, explainable, continual-learning Transformer that jointly delivers high detection accuracy, lifelong adaptability, and operator-facing interpretability for 5G/6G slicing. Methodologically, a multi-head self-attention encoder models short- and long-range dependencies in high-dimensional slice telemetry; Elastic Weight Consolidation (EWC) and an experience-replay buffer mitigate catastrophic forgetting across a defined sequence of attack-introduction tasks; and an explainability layer combines attention heatmaps with SHAP attributions. Three complementary variants (Transformer–LSTM, Graph Attention, and Autoencoder–Transformer ensemble) address diverse anomaly types. On 5G-core and Open-RAN datasets, the framework attains 98.87% accuracy, 98.90% precision, 98.70% recall, and 1.8 ms/sample latency, outperforming LSTM, CNN, and Random Forest baselines (McNemar p &amp;lt; 0.001). Continual-learning evaluation across five sequential tasks yields a low forgetting measure (FM = 0.013) versus 0.212 for naive fine-tuning, and ablation studies isolate the contribution of EWC, replay, and sequence length. Per-slice and per-attack analyses reveal where residual errors concentrate. These results show that integrating Transformer encoding, continual learning, and interpretability produces an accurate, adaptable, and transparent detector suited to live 5G/6G deployment.</dc:description>
	<dc:publisher xml:lang="en-US">Universitas Muhammadiyah Yogyakarta</dc:publisher>
	<dc:date>2026-07-01</dc:date>
	<dc:type>info:eu-repo/semantics/article</dc:type>
	<dc:type>info:eu-repo/semantics/publishedVersion</dc:type>
	<dc:format>application/pdf</dc:format>
	<dc:identifier>https://journal.umy.ac.id/index.php/jrc/article/view/28510</dc:identifier>
	<dc:identifier>10.18196/jrc.v7i4.28510</dc:identifier>
	<dc:source xml:lang="en-US">Journal of Robotics and Control (JRC); Vol. 7 No. 4 (2026); 4067-4083</dc:source>
	<dc:source>2715-5072</dc:source>
	<dc:source>2715-5056</dc:source>
	<dc:language>eng</dc:language>
	<dc:relation>https://journal.umy.ac.id/index.php/jrc/article/view/28510/12805</dc:relation>
	<dc:rights xml:lang="en-US">Copyright (c) 2026 Zainab J. Taher, Maha I. Raheem, Sinan Q. Salih, Poh Soon JosepNg, Taha Almulaisi, Felix Pasila, Koo Yuen Phan, Hilal A. Fadhil</dc:rights>
	<dc:rights xml:lang="en-US">https://creativecommons.org/licenses/by-sa/4.0</dc:rights>
</oai_dc:dc>