Systems and methods for fault detection based on peer statistics

Abstract

While many network components include diagnostic capabilities that are sometimes implemented at the hardware level, these diagnostics can be unreliable. Thus, false indications of operability or inoperability can result when these diagnostics are relied upon exclusively. To better detect operational problems with a network component, operational parameters from the network component and one or more peer devices are analyzed to determine whether the network component is operational. In some embodiments, data derived from these operational parameters is provided to a supervised machine learning model, and the model provides output indicating the operational status of the network component. Some embodiments binarize operational parameters of a device and compute a maximum duration the binarized parameters indicate inactivity of the device. Some embodiments compute a moving average of the binarized parameters. The maximum duration and/or moving average(s) are provided to the machine learning model in some embodiments.