A Semi-Supervised Autoencoder Approach for Efficient Intrusion Detection in Network Traffic
Arthur da Costa Almeida, Luís Gonçalves, Cleber Zanchettin, Byron Leite Dantas Bezerra · 2023
Intrusion detection systems (IDS) play a crucial role in protecting computer networks against cyber-attacks. An IDS typically recognizes known intrusion patterns or identifies unusual user behavior. With advancements in machine learning and deep neural networks, efficient intrusion detection solutions have been developed. In this paper, we present an autoencoder-based solution trained in a semi-supervised manner to detect anomalies in network traffic. The autoencoder is trained on normal network traffic, allowing the model to learn a compact representation of the regular traffic data. The model uses the reconstruction error as a mechanism for identifying anomalies in the network traffic. The proposed solution is designed to be efficient in terms of computational resources. The experiments were performed using the NSL-KDD dataset, a widely used benchmark for intrusion detection systems, and achieved an accuracy of 90.49% and an Fl-score of 91.81%, outperforming state-of-the-art methods in the same dataset.