Generative Replay Menggunakan Gans Untuk Continual Learning Pada Sistem Deteksi Intrusi Jaringan

Dzakwan, Roofiif Alria (2026) Generative Replay Menggunakan Gans Untuk Continual Learning Pada Sistem Deteksi Intrusi Jaringan. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025221012-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Sistem Deteksi Intrusi Jaringan (NIDS) berbasis Deep Learning sering menghadapi masalah catastrophic forgetting, di mana kemampuan mendeteksi serangan lama menurun drastis saat mempelajari ancaman baru. Selain itu, metode konvensional yang menyimpan seluruh data historis untuk pelatihan ulang membebani memori dan penyimpananPenelitian ini mengusulkan framework Continual Learning berbasis Generative Replay yang mengintegrasikan Convolutional Neural Network (CNN), Replay Buffer, Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP), dan Proximal Policy Optimization (PPO). CNN digunakan sebagai model klasifikasi utama, Replay Buffer berfungsi mempertahankan informasi dari pembelajaran sebelumnya, WGAN-GP menghasilkan data sintetis yang merepresentasikan pola serangan terdahulu, sedangkan PPO berperan sebagai agen adaptif dalam menentukan strategi replay. Evaluasi dilakukan menggunakan metrik Accuracy, Precision, Recall, Macro F1-Score, Balanced Accuracy, dan absolute forgetting. Hasil eksperimen menunjukkan bahwa Replay Buffer memberikan kontribusi terbesar dalam mempertahankan performa model selama pembelajaran bertahap. Selain itu, visualisasi Principal Component Analysis (PCA) menunjukkan bahwa WGAN-GP mampu menghasilkan data sintetis yang mengikuti pola distribusi data asli. Hasil penelitian menunjukkan bahwa mekanisme Generative Replay dapat membantu mengurangi dampak catastrophic forgetting pada NIDS berbasis Continual Learning.
=======================================================================================================================================
With the increasing use of computer networks, cyber threats have become more diverse, requiring Network Intrusion Detection Systems (NIDS) to continuously adapt to emerging attacks without losing previously acquired knowledge. This challenge is known as catastrophic forgetting in Continual Learning. This study proposes a Generative Replay based framework that integrates a Convolutional Neural Network (CNN), Replay Buffer, Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP), and Proximal Policy Optimization (PPO). The proposed framework is evaluated using four NIDS datasets organized into four sequential datasets based on Accuracy, Precision, Recall, Macro F1-Score, Balanced Accuracy, and Absolute Forgetting. Experimental results show that the Full Scenario achieves the best performance with an average Macro F1-Score of 72.40%, while Principal Component Analysis (PCA) visualization demonstrates that WGAN-GP generates synthetic data with a distribution closely resembling the original data. These findings indicate that the proposed framework has been successfully implemented and achieves the best performance under the experimental configuration employed in this study.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Network Intrusion Detection System (NIDS), Continual Learning, Catastrophic Forgetting.
Subjects:	T Technology > T Technology (General) > T57.5 Data Processing
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Roofiif Alria Dzakwan
Date Deposited:	24 Jul 2026 03:39
Last Modified:	24 Jul 2026 03:39
URI:	http://repository.its.ac.id/id/eprint/137683

Actions (login required)

View Item