Electronic Theses and Dissertation

Keamanan siber menjadi isu krusial di tengah pesatnya perkembangan digital, seiring meningkatnya volume lalu lintas jaringan dan kompleksitas serangan siber. Studi ini bertujuan untuk membandingkan performa model Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), dan dua model hybrid (CNN-LSTM dan LSTM-CNN) dalam mendeteksi serangan siber pada lalu lintas jaringan. Dataset yang digunakan adalah HIKARI-2021, yang mencakup berbagai jenis lalu lintas jaringan, baik normal maupun berbahaya. Penelitian ini melalui beberapa tahapan, termasuk analisis deskriptif, pengolahan data, serta pemodelan dengan deep learning. Model CNN digunakan untuk mengekstraksi fitur spasial dari data lalu lintas jaringan, sedangkan LSTM diterapkan untuk menangkap pola temporal. Pendekatan hybrid menggabungkan keunggulan kedua model untuk meningkatkan akurasi klasifikasi. Hasil eksperimen menunjukkan bahwa model hybrid LSTM-CNN memberikan performa terbaik dengan akurasi sebesar 93,30%, precision sebesar 93,22%, recall sebesar 93,30%, dan F1-score sebesar 93,20%. Selain itu, penerapan SMOTE terbukti meningkatkan akurasi model secara signifikan, yakni sebesar 10–12% dibandingkan sebelum dilakukan penyeimbangan data. Temuan ini menegaskan bahwa penggunaan teknik hybrid deep learning yang dikombinasikan dengan strategi penyeimbangan data dapat secara efektif meningkatkan akurasi klasifikasi dan deteksi serangan siber pada sistem keamanan jaringan. Implementasi metode ini diharapkan dapat memperkuat sistem keamanan siber yang lebih adaptif dan responsif terhadap ancaman siber di masa depan.
Kata kunci: keamanan siber, deep learning, CNN, LSTM, model hybrid, ketidakseimbangan data

Cybersecurity has emerged as a critical issue amid the rapid advancement of digital technologies, driven by the increasing volume of network traffic and the growing complexity of cyberattacks. This study aims to compare the performance of Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and two hybrid models (CNN-LSTM and LSTM-CNN) in detecting cyber threats within network traffic. The dataset used is HIKARI-2021, which includes various types of network traffic, encompassing both benign and malicious activities. The research process involved several stages, including descriptive analysis, data preprocessing, data balancing using the Synthetic Minority Over-sampling Technique (SMOTE), and deep learning-based modeling. CNN was employed to extract spatial features from network traffic data, while LSTM was utilized to capture temporal patterns. The hybrid approach combines the strengths of both models to enhance classification accuracy. Experimental results show that the LSTM-CNN hybrid model achieved the best performance, with an accuracy of 93.30%, a precision of 93.22%, a recall of 93.30%, and an F1-score of 93.20%. Furthermore, the application of SMOTE significantly improved model performance, increasing accuracy by approximately 10–12% compared to the imbalanced dataset. These findings demonstrate that the combination of hybrid deep learning techniques and data balancing strategies can effectively improve classification accuracy and cyberattack detection in network security systems. The implementation of this approach is expected to strengthen cybersecurity systems that are more adaptive and responsive to evolving threats. Keywords: cybersecurity, deep learning, CNN, LSTM, hybrid model, data imbalance