Fath, Nadira Milha Nailul (2025) Deteksi Serangan pada Jaringan Komputer Menggunakan Anomaly Transformer pada Data Deret Waktu. Other thesis, Institut Teknologi Sepuluh Nopember.
![]() |
Text
5025211253-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy |
Abstract
Deteksi anomali pada data deret waktu merupakan proses penting yang bertujuan untuk mengidentifikasi kejadian-kejadian yang menyimpang dari pola normal dalam suatu sistem, dengan menganalisis tren historis yang terjadi. Proses ini sangat krusial dalam berbagai domain seperti pemantauan jaringan, keuangan, industri manufaktur, hingga kesehatan, karena mampu memberikan peringatan dini terhadap potensi gangguan atau kerusakan. Dalam konteks pemantauan jaringan, data yang digunakan berupa network log, di mana anomali dapat muncul sebagai lonjakan aktivitas yang tidak biasa dalam kurun waktu tertentu. Tantangan utama terletak pada kompleksitas pola fitur secara global serta pola temporal antar langkah waktu. Penelitian ini berfokus pada penerapan dan perbandingan tiga model deteksi anomali berbasis deep learning, yaitu Anomaly Transformer, TranAD, serta model Donut yang telah dimodifikasi dari univariate menjadi multivariate. Dataset yang digunakan meliputi OS Scan, Active Wiretap, Fuzzing, dan ARP MitM, yang merupakan representasi numerik dari lalu lintas jaringan nyata dengan distribusi kelas yang sangat tidak seimbang. Hasil penelitian pada Anomaly Transformer menunjukkan peningkatan signifikan setelah dilakukan hyperparameter tuning terhadap parameter k dan anormly_ratio. Pada dataset OS Scan, f1-score meningkat dari 0,0549 menjadi 0,4167, mengalami kenaikan 659%. Dataset lain seperti Fuzzing, ARP MitM, dan Active Wiretap juga menunjukkan peningkatan masing-masing 21,83%, 9,16%, dan 14,38%. Model ini juga menunjukkan performa lebih konsisten dibandingkan Donut dan TranAD, dengan precision stabil di atas 0,7000 pada semua konfigurasi. Penelitian ini memberikan kontribusi terhadap pengembangan sistem deteksi anomali yang lebih adaptif dan akurat pada data deret waktu jaringan yang kompleks.
=======================================================================================================================================
Anomaly detection in time series data is an important process aimed at identifying events that deviate from the normal pattern within a system by analyzing the historical trends that occur. This process is crucial in various domains such as network monitoring, finance, manufacturing industry, and healthcare, as it can provide early warnings of potential disruptions or damages. In the context of network monitoring, the data used consists of network logs, where anomalies can appear as spikes in unusual activity over a certain period. The main challenge lies in the complexity of global feature patterns as well as temporal patterns across time steps. This research focuses on the application and comparison of three deep learning-based anomaly detection models, namely Anomaly Transformer, TranAD, and the Donut model, which has been modified from univariate to multivariate. The datasets used include OS Scan, Active Wiretap, Fuzzing, and ARP MitM, which are numerical representations of real network traffic with highly imbalanced class distributions. The research results on the Anomaly Transformer show significant improvement after hyperparameter tuning of the parameters k and anomaly_ratio. In the OS Scan dataset, the f1-score increased from 0.0549 to 0.4167, a 659% increase. Other datasets such as Fuzzing, ARP MitM, and Active Wiretap also showed improvements of 21.83%, 9.16%, and 14.38%, respectively. This model also demonstrated more consistent performance compared to Donut and TranAD, with precision stable above 0.7000 in all configurations. This research contributes to the development of more adaptive and accurate anomaly detection systems for complex network time series data.
Item Type: | Thesis (Other) |
---|---|
Uncontrolled Keywords: | Anomaly Detection, Time series, Transformer, Unsupervised Learning, Anomaly Detection, Time series, Transformer, Unsupervised Learning |
Subjects: | T Technology > T Technology (General) |
Divisions: | Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis |
Depositing User: | Nadira Milha Nailul Fath |
Date Deposited: | 28 Jul 2025 09:30 |
Last Modified: | 28 Jul 2025 09:30 |
URI: | http://repository.its.ac.id/id/eprint/122260 |
Actions (login required)
![]() |
View Item |