The Inference Engine is responsible for detecting anomalies and inferring the root cause using above mentioned diagnostic metrics. In detail, when the application data throughput decreases, the Inference Engine is triggered to localize isolated nodes in the network. Here, a node is diagnosed as isolated if the symptom persists longer than the learned 90th percentile of its Disconnection Duration. If the throughput from several sensor nodes decreases at the same time, the Inference Engine tries to locate a potentially faulty link including potentially affected nodes. If the throughput of all nodes dropped at once, there is likely a failure at the sink.