植基於機器學習之分散式阻斷服務攻擊偵測

現今物聯網的快速成長，在我們享受科技帶來的便利時，分散式阻斷服務攻擊（Distributed Denial of Service, DDoS）一直是伴隨在暗處的隱憂，伺服器的資訊安全受到威脅，使用者的服務受到阻斷，公司面臨財政、名聲上的損失。隨著人工智慧的興起，人工智慧技術也被應用於DDoS入侵偵測系統中，本研究比較常見機器學習的方法，應用於DDoS入侵偵測系統當中，以比較哪種機器學習的方法較適用於現今的DDoS入侵樣態。而深度學習近幾年逐漸興起，有足夠的資料就可以使電腦自我學習特徵並進行判斷，本研究利用卷積神經網路與長短期記憶網路，探究機器學習和深度學習應用在DDoS入侵檢測上的效能表現。而為降低訓練的複雜度與訓練的時間，本研究以隨機森林（Random Forest）、交叉驗證之遞歸特徵消除（Recursive Feature Elimination with Cross-validation, RFECV）和主成分分析（Principal Component Analysis, PCA）進行特徵的選擇，以探究哪種特徵選取的方法適合用於機器學習與深度學習，比較其中準確率與效能的差異。

關鍵字

分散式服務阻斷攻擊；機器學習；深度學習

並列摘要

Nowadays, with the rapid growth of the Internet, the Distributed Denial of Service (DDoS) attack has always been a concern. The information security of the server is threatened and the company faces financial losses. With the rise of artificial intelligence, artificial intelligence technology has also been applied to DDoS intrusion detection systems. As the growth of Artificial Intelligence which technology has also been applied to DDoS intrusion detection systems. This study compares common machine learning methods and applies them to DDoS intrusion detection systems to compare which machine learning method is more suitable for today's DDoS intrusion patterns. Deep learning has gradually emerged in recent years. Computer self-learning features can be judged with sufficient information. This study uses Convolutional Neural Networks and Long Short-Term Memory Networks to explore machine learning and deep learning's performance which applications in DDoS intrusion detection. In order to reduce the complexity of training and training time, this study uses Random Forest, Recursive Feature Elimination with Cross-validation (RFECV) and Principal Component Analysis (PCA) to select the feature. For explore which feature selection method is suitable for machine learning and deep learning, and compare the difference between accuracy and performance.

並列關鍵字

Distributed Denial of Service attack ； Machine Learning ； Deep Learning

國際替代計量

植基於機器學習之分散式阻斷服務攻擊偵測

全文下載

主題瀏覽