Pengembangan Model Machine Learning untuk Deteksi Serangan Siber

Abstract

As cyber attacks become more complex, conventional detection systems such as signature-based Intrusion Detection Systems (IDS) are starting to experience limitations in accurately recognizing new threats. This study aims to develop a deep learning-based machine learning model to detect cyber attacks by combining the Convolutional Neural Network (CNN) architecture as feature extraction and Random Forest (RF) as classification. The dataset used in this study is CICIDS2018, which has several types of network attacks. The study was conducted in two Incremental stages, namely model development and model implementation into a web application that can detect cyber attacks using input from .pcap files and output in .csv format. The evaluation results showed that the combination of the CNN model as feature extraction and RF as classification obtained an accuracy of 91%, making this model combination better than the combination of the CNN-XGBoost model which only obtained an accuracy of 89%. This model also proved to be better than the single CNN model which only obtained an accuracy of 87% and better than the previous research architecture which obtained an accuracy of 90%. The results of all tested models can be concluded that using a combination of CNN-RF models produces better accuracy results.