Sistem Deteksi Instrusi Jaringan Adaptif Berbasis Self-supervised Continual Learning

Rabbani, Fauzan Dzikri (2026) Sistem Deteksi Instrusi Jaringan Adaptif Berbasis Self-supervised Continual Learning. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025221311-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Server web modern dan sistem kontrol sangat bergantung pada efisiensi Sistem Deteksi Intrusi (IDS) untuk mitigasi serangan yang dapat mengganggu kinerja atau privasi. Meskipun banyak studi telah mengeksplorasi IDS berbasis Machine Learning (ML-IDS), model tradisional sering mengalami catastrophic forgetting selama pembelajaran berkelanjutan, menyebabkan mereka kehilangan kemampuan untuk mendeteksi pola serangan yang sudah dikenal saat beradaptasi dengan ancaman baru atau zero-day. Untuk mengatasi hal ini, studi ini bertujuan untuk mengimplementasikan dan mengevaluasi efektivitas kerangka kerja Continual Learning yang dirancang untuk mengurangi catastrophic forgetting. Kerangka kerja ini mengintegrasikan konverter data tabular ke gambar, convolutional masked autoencoder untuk ekstraksi fitur laten, detektor novelty, dan memori hierarkikal. Evaluasi dilakukan menggunakan dataset X-IIoTID dan CIC-IDS-2017. Hasil pengujian menunjukkan bahwa kerangka kerja ini efektif memitigasi catastrophic forgetting, dibuktikan dengan perolehan nilai Backward Transfer (BWT) yang positif sebesar 0,0499 pada X-IIoTID dan 0,1163 pada CIC-IDS-2017 dengan konfigurasi optimal pada ukuran citra 8x8 dan kapasitas memori 5000 sampel. Selain itu, sistem mencapai nilai rata-rata gabungan Lifelong PR-AUC (LL-PR AUC) sebesar 0,6189, mengungguli metode continual learning terdahulu yang dibandingkan.
=========================================================================================================================================
Modern webservers and control systems rely heavily on the efficiency of Intrusion Detection Systems (IDS) to mitigate attacks that could compromise performance or privacy. While numerous studies have explored Machine Learning-based IDS (ML-IDS), traditional models often suffer from catastrophic forgetting during continuous learning, causing them to lose the ability to detect established attack patterns when adapting to new or zero-day threats. To address this, this study aims to implement and evaluate the effectiveness of a Continual Learning framework designed to mitigate catastrophic forgetting. The framework integrates a tabular-to-image data converter, a convolutional masked autoencoder for latent feature extraction, a novelty detector, and hierarchical memory. The evaluation is conducted using the X-IIoTID and CIC-IDS-2017 datasets. Experimental results indicate that the framework effectively mitigates catastrophic forgetting, as evidenced by positive Backward Transfer (BWT) values of 0.0499 on X-IIoTID and 0.1163 on CIC-IDS-2017, with an optimal configuration of 8x8 image size and a memory capacity of 5000 samples. Furthermore, the system achieves a combined average Lifelong PR-AUC (LL-PR AUC) of 0.6189, outperforming previous continual learning methods compared in this research.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Continual Learning, Masked Autoencoder, Intrusion Detection, Catastrophic Forgetting, Tabular-to-image,
Subjects:	T Technology > T Technology (General) > T174.5 Technology--Risk assessment. T Technology > T Technology (General) > T57.5 Data Processing T Technology > T Technology (General) > T57.8 Nonlinear programming. Support vector machine. Wavelets. Hidden Markov models. T Technology > T Technology (General) > T58.5 Information technology. IT--Auditing
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Fauzan Dzikri Rabbani
Date Deposited:	24 Jul 2026 07:50
Last Modified:	24 Jul 2026 07:50
URI:	http://repository.its.ac.id/id/eprint/137777

Actions (login required)

View Item