14 February 2025 in IEEE Internet of Things Journal
Due to the difficulty in predefining novel attack patterns and the scarcity of sufficient malicious training samples in specialized Internet of Things (IoT) scenarios, the focus of IoT intrusion detection researchers has shifted toward anomaly-based techniques like Autoencoder. These techniques detect attacks based on the degree of deviation and rely minimally on malicious samples. However, current machine learning (ML) and deep learning (DL) implementations lack explainability. Although some methods provide post-hoc interpretations, they are limited and partial, failing to reflect the entire decision-making process. To address this challenge, we investigate an explainable anomaly-based intrusion detection system (IDS) that translates the inference process of the Autoencoder into the high-fidelity allow-list rule library, thereby balancing the detection capability and interpretability. First, we assume that benign traffic follows a complex global distribution composed of several irrelevant local distributions. The clustering algorithm is performed in an extended feature space consisting of reconstruction loss and embeddings to decompose local distributions. Then, we deploy an approach based on Gradient Ascent to explore the boundary rules of each local distribution. The allow-list rule library that reflects Autoencoder’s inference process can be constructed by merging these boundary rules. Comprehensive evaluation experiments demonstrate that the extracted allow-list rule library accurately reproduces Autoencoder’s inference process and effectively detects IoT intrusions.
17 February 2025 in IEEE Internet of Things Journal
Quantifying the deviation of testing samples from known benign traffic in the form of novelty scores is essential for identifying new malicious traffic and detecting concept drifts of benign traffic. Most existing solutions generate novelty scores based on the outputs from the supervised neural network’s last layers. However, the performance of these supervised techniques is significantly compromised in the unbalanced dynamic interconnected specialized IoT network traffic such as Electronic Power Grid. To solve this challenge, we investigate an improved novelty score calculation approach. It employs the targeted distance loss to ensure that different known classes form class-specific, dense clusters within the embedding space outputted by the neural network. The minimum distance between the testing sample and the centers of each known class is utilized as the novelty score. Extensive comparison experiments on public datasets and real Electronic Power Grid traffic demonstrate that our approach outperforms existing supervised techniques. As the imbalance in known class distributions increases, our proposed method consistently achieves higher AUROC and AUPR than baselines, with a reduced FPR95. This study analyzes imbalanced known class distributions’ negative impact on novelty detection and provides practical improvement.