Machine learning covers algorithms that fit a model to data and then use that model to classify, predict or group new observations. In IoT it is the main way to extract meaning from streams of telemetry that are too large or too subtle for hand-written thresholds. A model can learn what a healthy motor sounds like from weeks of recordings, then flag departures from that baseline.
The two phases are model training, which is computationally heavy and usually done in the cloud, and inference, which can run on constrained hardware. The trade-off is that a model is only as good as its training data; a sensor moved to a new site may behave differently and need retraining. See What is AIoT? for how this fits with connected devices.
Part of the IoT Glossary. Spotted an error or a missing term? Tell us.