人工智慧方法對於網路入侵攻擊的預測

Artificial Intelligence methods for the prediction of network intrusion attacks

郭芳瑜(Fun Yu Kuo)；林宗儀(Tsung-I Lin)

人工智慧 ； 深度神經網路 ； 卷積神經網路 ； 網路入侵攻擊 ； Artificial Intelligence ； Convolutional Neural Networks ； Deep Neural Networks ； Network Intrusion Attacks

智慧科技與應用統計學報

22卷1期（2024 / 07 / 01）

1 - 25

繁體中文;英文

本研究探索了人工智慧（Artificial Intelligence, AI）技術在預測網路入侵攻擊方面的應用，以先進的算法處理大量資訊，進而增加網路安全性並大幅提升傳統方法的性能。採用UNSW-NB15資料集作為研究對象，該資料涵蓋從偵查攻擊到系統漏洞利用的多種攻擊情境，並通過特徵分析評估各特徵間的相關性與其重要性。我們選擇多種機器學習（決策樹、極限梯度提升、隨機森林）和深度學習（深度神經網路、卷積神經網路）模型進行研究。實驗結果顯示，卷積神經網路（Convolutional Neural Network, CNN）在未執行大量資料前處理的情況下，具有最高的準確率（85.58%）和精確率（79.52%），表示其預測效果優異，對於網路攻擊和非網路攻擊能夠有效辨別。因此，通過本文將證實AI方法運用於網路入侵攻擊預測中的潛在價值，並為未來相關研究提供基礎。

This research explores the application of artificial intelligence(AI) technology in predicting network intrusion attacks, employing advanced algorithms to process large amounts of information, thereby enhancing network security and significantly improving the performance of traditional methods. The UNSW-NB15 dataset is utilized for the study, covering various attack scenarios from reconnaissance attacks to system vulnerability exploits, and assessing the correlation and importance of features through feature analysis. We select multiple machine learning (decision trees, extreme gradient boosting(XGBoost), random forests) and deep learning (deep neural networks(DNNs), convolutional neural networks(CNNs)) models for research. Experimental results show that CNN achieves the highest accuracy (85.58%) and precision (79.52%) without extensive data preprocessing, indicating its excellent predictive performance in effectively distinguishing between network attacks and non-network attacks. Therefore, this study confirms the potential value of AI methods in predicting network intrusion attacks and provides a foundation for future related research.