Application of machine learning for building predictive models enabling smart fail over between different network media types

摘要

Computing devices are configured to passively monitor network stacks and protocols for a respective computing device, transmit metadata and statistics gathered by the monitoring to a remote service, and utilize a crowd-sourced heuristic model responsively generated by the remote service to proactively predict connectivity issues and connect to a best available network media and access device for the network media. A computing device's operating system may monitor various networking protocols without the computing device engaging in constant network activities (e.g., video streaming). The statistics obtained from this passive monitoring can be utilized by the remote service using various machine learning techniques to predict when networks will subsequently fail. Profiles are developed and sorted within the model to be used by individual computing devices to seamlessly connect to access devices based on performance, as opposed to connecting to the access device previously utilized by the user.