Anomaly detection/prediction is the first step to secure IoT systems. It usually relies on wide domain knowledge to build up the tools to automatically detect/predict abnormal events or behaviors of an IoT system. However, an IoT system may consist of machines with different capabilities, functionalities and ages. Furthermore, abnormal events or behaviors are usually rare events. It is time-consuming and high-cost to build up the domain knowhow of the IoT systems and collect enough data points of the anomaly. In this article, we first identify the issues and challenges. Then we illustrate a general environment for anomaly detection/prediction on the IoT systems. Then we survey the core technologies and existing solutions that may be applied for anomaly detection/prediction. We also identify what cannot be achieved by the existing solutions. Then considering four datasets, we show the performance comparison for different solutions by running experiments.

中文翻译：

---

物联网的异常检测/预测：最新技术和未来

异常检测/预测是保护物联网系统的第一步。它通常依赖于广泛领域的知识来构建工具，以自动检测/预测IoT系统的异常事件或行为。但是，物联网系统可能由具有不同功能，功能和年龄的机器组成。此外，异常事件或行为通常是罕见事件。建立物联网系统的领域知识并收集足够的异常数据点是耗时且高成本的。在本文中，我们首先确定问题和挑战。然后，我们说明了用于IoT系统上异常检测/预测的一般环境。然后，我们调查了可用于异常检测/预测的核心技术和现有解决方案。我们还确定了现有解决方案无法实现的目标。