A Survey on Explainable Anomaly Detection for Industrial Internet of Things
Zijie Huang, Yulei Wu
Abstract
Anomaly detection techniques in the Industrial Internet of Things (IIoT) are driving traditional industries towards an unprecedented level of efficiency, productivity and performance. They are typically developed based on supervised and unsupervised machine learning models. However, some machine learning models are facing “black box” problems, namely the rationale behind the algorithm is not understandable. Recently, several models on explainable anomaly detection have emerged. The “black box” problems have been studied by using such models. But few works focus on applications in the IIoT field, and there is no related review of explainable anomaly detection techniques. In this survey, we provide an overview of explainable anomaly detection techniques in IIoT. We propose a new taxonomy to classify the state-of-the-art explainable anomaly detection techniques into two categories, namely intrinsic based explainable anomaly detection and explainer based explainable anomaly detection. We further discuss the applications of explainable anomaly detection across various IIoT fields. Finally, we suggest future study options in this rapidly expanding subject.