An anomaly is anything that deviates from the expected or normal outcomes. Detecting anomalies can be important in Industrial Process Monitoring (IPM), where data-driven fault detection and diagnosis can help achieve achieve higher levels of safety, efficiency, and quality.
In this recipe, we'll look at methods for outlier detection. We'll go through an example of outlier detection in a time series with Python Outlier Detection (pyOD), a toolbox for outlier detection that implements many state-of-the-art methods and visualizations. PyOD's documentation can be found at https://pyod.readthedocs.io/en/latest/.
We'll apply an autoencoder for a similarity-based approach, and then an online learning approach suitable for finding events in streams of data.
Getting ready
This recipe will focus on finding outliers. We'll demonstrate how to do this with the pyOD library including an autoencoder approach. We'll also outline the upsides and downsides...