Autoencoder-based Temporal Convolutional Network for Real-Time Anomaly Detection in Large-Scale Network Traffic

Nowadays, the rapid growth of network traffic has needed the development of efficient anomaly detection systems to ensure network security and reliability. However, the existing One-Class Support Vector Machine (One-class SVM) model often relay on manual feature engineering which was led to time-consuming and ineffective in detecting unknown threats. Hence, this research proposes an Autoencoder-based Temporal Convolutional Network (AE-TCN) for real-time anomaly detection in large-scale network traffic data. This process begins with the collection of input data from Coburg Intrusion Detection Data Set (CIDDS-001) dataset which consists of real time network traffic data from an internal server and external server. Then, the input data is preprocessed with the help of one-hot encoding to convert the categorical variables into numerical variables. After that, Recursive Feature Elimination (RFE) is employed to select the features namely class label, packets and type of attack by eliminating the irrelevant features from the input data. Finally, AE-TCN is introduced to identify the anomalies in large-scale network traffic data from reconstructed nominal patterns. From the results, the proposed AE-TCN achieved better results in terms of accuracy (99.98%), precision (99.95%), recall (99.26%), and F1 score (99.63%) respectively when compared to existing Convolutional Neural Network (CNN) model.