Using autoencoders to detect anomalies in labeled data

If you have labeled data, you can train a model to detect whether the data is normal or abnormal. For example, reading the current of an electric motor can show when extra drag is put on the motor by such things as failing ball bearings or other failing hardware. In IoT, anomalies can be a previously known phenomenon or a new event that has not been seen before. As the name suggests, autoencoders take in data and encode it to an output. With anomaly detection, we see whether a model can determine whether data is non-anomalous. In this recipe, we are going to use a Python object detection library called pyod.