SISTEM DETEKSI SERANGAN LOW AND SLOW DDOS MENGGUNAKAN DEEP LEARNING PADA APPLICATION LAYER

Ganjar Muhammad Parikesit, Dr. Lukman Heryawan, S.T., M.T.

2024 | Skripsi | ILMU KOMPUTER

Dalam era teknologi informasi yang maju, serangan Distributed-Denial-of- Service (DDoS) khususnya jenis low and slow DDoS, menjadi ancaman serius terhadap ketersediaan layanan pada website. Penelitian ini mengusulkan penggunaan algoritma deep learning untuk mendeteksi serangan low and slow DDoS pada application layer, sebuah pendekatan yang belum banyak diteliti sebelumnya. Dengan fokus pada analisis log traffic protokol HTTP, penelitian ini mengembangkan sistem berbasis middleware yang mampu mengidentifikasi client yang melakukan serangan low and slow DDoS. Implementasi sistem ini memungkinkan pembatasan akses terhadap client terduga serangan, sehingga meningkatkan ketersediaan layanan.
Melalui penelitian ini, ditemukan bahwa algoritma deep learning, khususnya model Long Short-Term Memory (LSTM) dengan panjang input 8, menunjukkan kinerja yang sangat baik dengan akurasi deteksi mencapai 99.18?n waktu prediksi 4.912ms. Hasil ini menunjukkan efektivitas deep learning dalam mengenali pola serangan low and slow DDoS pada application layer. Selain itu, implementasi sistem deteksi yang dikembangkan berhasil mengurangi penggunaan sumber daya CPU dan memori, serta jumlah koneksi aktif, menandakan sistem ini efektif dalam mitigasi dampak serangan low and slow DDoS dan meningkatkan kinerja serta keamanan server.

In the advanced era of information technology, Distributed-Denial-of- Service (DDoS) attacks, especially the low and slow DDoS type, pose a serious threat to the availability of services on websites. This study proposes the use of deep learning algorithms to detect low and slow DDoS attacks at the application layer, an approach that has not been extensively researched previously. Focusing on the analysis of HTTP protocol traffic logs, this research develops a system based on a middleware capable of identifying clients conducting low and slow DDoS attacks. The implementation of this system enables the restriction of access to suspected attacking clients, thereby enhancing service availability.
Through this research, it was found that deep learning algorithms, particularly the Long Short-Term Memory (LSTM) model with an input length of 8, demonstrated excellent performance with a detection accuracy reaching 99.18% and a prediction time of 4.912ms. These results indicate the effectiveness of deep learning in recognizing low and slow DDoS attack patterns at the application layer. Additionally, the implementation of the developed detection system successfully reduced CPU and memory resource usage, as well as the number of active connections, indicating its effectiveness in mitigating the impact of low and slow DDoS attacks and improving server performance and security.

Kata Kunci : Serangan DDoS, Low and slow DDoS, Algoritma deep learning, Application layer