Anomaly detection models the normal behavior of a signal and raises a flag when new data falls outside it. Methods range from simple statistics, such as a rolling mean and standard deviation, to autoencoders and clustering trained on healthy operation. The strength for IoT is that failures are rare and varied, so labeled fault data is scarce, but healthy data is abundant.
A typical deployment watches time-series data from a pump and alerts when vibration and temperature drift together in an unfamiliar way. The weakness is false positives: a legitimate change in operating mode looks like an anomaly until the model learns it, and model drift can quietly erode accuracy. It is a core building block of predictive maintenance and of security monitoring on device fleets.
Part of the IoT Glossary. Spotted an error or a missing term? Tell us.