| 摘要: | 隨著網路流量日益複雜,現有入侵偵測系統(Intrusion Detection System, IDS)在面對未知與零時差攻擊時,主要挑戰在於難以同時捕捉網路資料中的局部靜態特徵(空間相關性)與長時間動態特徵(時序依賴性),特別是對 CICIDS2017 資料集中稀有攻擊類別,其召回率普遍偏低。
為解決此問題,本研究提出一種結合卷積神經網路(Convolutional Neural Network, CNN)與時間卷積網路(Temporal Convolutional Network, TCN)之混合式深度學習架構。該架構以前端 CNN 作為靜態特徵提取器,後端 TCN 採用因果膨脹卷積(Dilated Causal Convolution)以捕捉長時序關聯,從而實現對異常流量的高精度辨識。
研究採用公開的 CICIDS2017 資料集,並與隨機森林(Random Forest, RF)、支持向量機(Support Vector Machine, SVM)、長短期記憶網路(Long Short-Term Memory, LSTM)及 CNN-LSTM 等多種基準模型進行比較。實驗結果顯示,所提出之 CNN–TCN 模型於整體分類效能上表現卓越,準確率(Accuracy)達 99.6%,F1-score 為 0.996,均明顯優於單一 CNN、TCN 與其他對照模型。
本研究之主要貢獻包括:
(1)整合 CNN 與 TCN 架構,兼具空間與時序特徵學習能力;
(2)顯著提升少數類別攻擊之偵測率與模型穩定性;
(3)提供一可並行化訓練之高效率 IDS 深度學習框架,具實務應用價值。
With the increasing complexity of network traffic, existing Intrusion Detection Systems (IDS) face significant challenges in effectively capturing both local static features (spatial correlations) and long-term dynamic dependencies (temporal rela-tions) within network data, especially when dealing with rare attack categories in the CICIDS2017 dataset, where recall rates are often low.
To address this issue, this study proposes a hybrid deep learning architecture that integrates a Convolutional Neural Network (CNN) and a Temporal Convolutional Network (TCN). In the proposed model, the CNN serves as a front-end feature extrac-tor for spatial characteristics, while the TCN employs dilated causal convolutions to capture long-range temporal dependencies, enabling precise recognition of anomalous traffic patterns.
The proposed approach is validated using the publicly available CICIDS2017 da-taset and compared against multiple baseline models, including Random Forest (RF), Support Vector Machine (SVM), Long Short-Term Memory (LSTM), and CNN-LSTM. Experimental results demonstrate that the CNN–TCN model achieves superior overall classification performance, with an accuracy of 99.6% and an F1-score of 0.996, out-performing standalone CNN, TCN, and other baseline models.
The main contributions of this study are as follows:
(1) Integration of CNN and TCN architectures to jointly learn spatial and temporal features;
(2) Significant improvement in the detection rate and stability for minority attack classes;
(3) A high-efficiency, parallelizable deep learning framework for IDS applications, offering strong practical applicability. |