Pendekatan Semi-Supervised Learning Untuk Sistem Deteksi Intrusi Jaringan Real-Time Berbasis GNN-AE Pada Arsitektur Structemp

Rinandri, Hafizh Athallah (2025) Pendekatan Semi-Supervised Learning Untuk Sistem Deteksi Intrusi Jaringan Real-Time Berbasis GNN-AE Pada Arsitektur Structemp. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5022211233-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Penelitian ini mengembangkan sistem deteksi intrusi jaringan berbasis Graph Neural Network Autoencoder (GNN-AE) yang terintegrasi dalam arsitektur Structemp-GNN dengan pendekatan semi-supervised learning untuk mendeteksi anomali pada trafik jaringan secara real-time. Model dirancang menggunakan arsitektur dual-path encoding yang memisahkan pemrosesan informasi struktural dan informasi temporal, yang kemudian akan digabungkan melalui fusi representasi laten yang dikontrol bobotnya oleh α . Kedua model lalu digabungkan menggunakan mekanisme fusi yang dikontrol bobotnya oleh β. Evaluasi dilakukan menggunakan enam dataset publik, dengan NF-UQ-NIDS-v2 sebagai dataset utama dan lima dataset tambahan untuk mengukur kemampuan generalisasi model. Hasil eksperimen menunjukkan bahwa model pada dataset NF-UQ-NIDS-v2 mencapai akurasi 97,51%, Presisi 99,34%, recall 97,51%, F1-score 97,53%, dan rata-rata waktu inferensi 10.07 mikrodetik per flow, menandakan efektivitas pendekatan ini dalam mendeteksi serangan jaringan secara adaptif, efisien, dan akurat dalam skenario operasional nyata.
==================================================================================================================================
This research develops a Graph Neural Network Autoencoder (GNN-AE) based Network Intrusion Detection System integrated in Structemp-GNN architecture with semi-supervised learning approach to detect anomalies in network trafik in real-time. The model is designed using a dual-path encoding architecture that separates the processing of structural information and temporal information, which will then be combined through the fusion of latent representations controlled by α . The two models are then combined using a weight-controlled fusion mechanism β. The evaluation is conducted using six public datasets, with NF-UQ-NIDS-v2 as the main dataset and five additional datasets to measure the generalization ability of the model. Experimental results show that the model on the NF-UQ-NIDS-v2 dataset achieves 97.51% accuracy, 99.34% precision, 97.51% recall, 97.53% F1-score, and an average inference time of 10.07 microseconds per flow, signifying the effectiveness of this approach in adaptively, efficiently, and accurately detecting network attacks in real operational scenarios.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Autoencoder, Deteksi Intrusi, Graph Neural Network, Real-Time Detection, Semi-supervised learning, Autoencoder, Graph Neural Network, Intrusion Detection, Real-Time Detection, Semi-supervised learning
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) Q Science > QA Mathematics > QA76.9.A25 Computer security. Digital forensic. Data encryption (Computer science) T Technology > T Technology (General)
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20201-(S1) Undergraduate Thesis
Depositing User:	Hafizh Athallah Rinandri
Date Deposited:	24 Jul 2025 07:34
Last Modified:	24 Jul 2025 07:34
URI:	http://repository.its.ac.id/id/eprint/121183

Actions (login required)

View Item