Chapter 11: Anomaly Detection – Predicting Failure with No Failure Examples
The goal of this chapter is to be able to predict exceptional events with only normal data at hand. Here, "normal" refers to frequent events, and "exceptional" refers to rare events, such as signals with a low and high amplitude. However, if we cannot provide sufficient examples of exceptional events to train a model that predicts them, we need to exclude all supervised learning techniques such as classification models. This means we can use only unsupervised learning techniques that predict normal behavior and determine the exceptional events based on those prediction errors. Such prediction of exceptional events is called anomaly detection.
In this chapter, we will show you various techniques for anomaly detection from Internet of Things (IoT) data. IoT refers to data signals from sensors that monitor a physical object, such as a mechanical rotor, a smart home, or the human...
