An Energy-efficient And Trustworthy Unsupervised Anomaly Detection Framework (EATU) for IIoT
Zijie Huang, Yulei Wu, Niccolò Tempini, Hui Lin, Hao Yin
Abstract
Many anomaly detection techniques have been adopted by Industrial Internet of Things (IIoT) for improving self-diagnosing efficiency and infrastructures security. However, they are usually associated with the issues of computational-hungry and “black box.” Thus, it becomes important to ensure that the detection is not only accurate but also energy-efficient and trustworthy. In this article, we propose an Energy-efficient And Trustworthy Unsupervised anomaly detection framework (EATU) for IIoT. The framework consists of two levels of feature extraction: (1) Autoencoder-based feature extraction and (2) Efficient DeepExplainer-based explainable feature selection. We propose an Efficient DeepExplainer model based on perturbation-focused sampling, which demonstrates the most computational efficiency among state-of-the-art explainable models. With the important features selected by Efficient DeepExplainer, the rationale of why an anomaly detection decision was made is given, enhancing the trustworthiness of the detection as well as improving the accuracy of anomaly detection. Three real-world IIoT datasets with high-dimensional features are used to validate the effectiveness of the proposed framework. Extensive experimental results demonstrate that in comparison with the state-of-the-art, our framework has the attributes of improved accuracy, trustworthiness (in terms of correctness and stability of the explanation), and energy-efficiency (in terms of wall-clock-time and resource usage).