Microsoft Corporation
IDENTIFYING PATTERNS IN EVENT LOGS TO PREDICT AND PREVENT CLOUD SERVICE OUTAGES

Abstract:

In non-limiting examples of the present disclosure, systems, methods and devices for predicting hardware failure events are presented. A time series comprising event log data for a plurality of events and a plurality of event types that occurred on a server computing device may be received. The time series may be filtered for a subset of the plurality of event types. The filtered time series may be processed with a recurrent neural network that has been trained to predict hardware failure events from time series data comprising the subset of the plurality of event types. A prediction may be made that a hardware failure event will occur on the server computing device within a threshold duration of time. A prophylactic follow-up action corresponding to the predicted hardware failure event may be performed.