Predicting and forecasting roaming issues in a wireless network

1. A method, comprising:

applying, by a network assurance service that monitors a network, labels to feature vectors of network characteristics associated with a plurality of Wi-Fi wireless access points in the network, wherein an applied label for a feature vector indicates whether a wireless access point associated with the feature vector experienced a threshold number of onboarding delays within a given time window, wherein an onboarding delay is determined by an amount of time exceeding a delay threshold, the amount of time indicative of when a client attempts to initially join the network or roams between wireless access points of the network;

training, by the network assurance service and based on the feature vectors and labels, a plurality of machine learning-based classifiers to predict onboarding delays;

using, by the network assurance service, one or more of the trained plurality of classifiers to predict onboarding delays for a particular wireless access point;

calculating, by the network assurance service, one or more classifier performance metrics for the one or more trained classifiers based on the predicted onboarding delays for the particular wireless access point and a history of onboarding delays; and

selecting, by the network assurance service, a particular one of the trained classifiers to monitor the network characteristics associated with the particular wireless access point, based on the one or more calculated classifier performance metrics.

2. The method as in claim 1 , further comprising:

using, by the network assurance service, the selected classifier to classify the network characteristics associated with the particular wireless access point, in order to predict onboarding delays for the particular wireless access point; and

initiating, by the network assurance service, a change to the monitored network to mitigate the predicted onboarding delays for the particular wireless access point.

3. The method as in claim 1 , wherein the one or more classifier performance metrics comprise at least one of: a classifier precision or a classifier recall.

4. The method as in claim 1 , wherein training, based on the feature vectors and labels, a plurality of machine learning-based classifiers to predict onboarding delays comprises:

generating a plurality of subsets of the wireless access points in the network, each subset comprising wireless access points that have similar network characteristics; and

training, for each subset of wireless access points, at least one machine learning-based classifier to predict onboarding delays using the feature vectors and labels associated with the wireless access points in the subset.

5. The method as in claim 4 , wherein generating the plurality of subsets of the wireless access points comprises:

using a machine learning-based clustering process on the network characteristics, to form the subsets of wireless access points.

6. The method as in claim 1 , wherein using one or more of the trained plurality of classifiers to predict onboarding delays for the particular wireless access point comprises:

obtaining a history of network characteristics associated with the particular wireless access point from a defined time period; and

using one or more of the plurality of classifiers on the history of network characteristics associated with the particular wireless access point, to predict onboarding delays for the defined period of time.

7. The method as in claim 6 , wherein calculating the one or more classifier performance metrics for the one or more trained classifiers based on the predicted onboarding delays for the particular wireless access point comprises:

obtaining a history of onboarding delays from the defined time period; and

comparing the history of onboarding delays from the defined time period to the predicted onboarding delays for the defined time period.

8. The method as in claim 1 , further comprising:

determining, by the network assurance service, that predicted onboarding delays for a second one of the wireless access points by one or more of the trained classifiers results in a performance metric for the one or more trained classifiers being below a threshold; and, in response

disable, by the network assurance service, the plurality of classifiers from classifying network characteristics of the second wireless access point for a period of time.

9. An apparatus, comprising:

one or more network interfaces to communicate with a network;

a processor coupled to the network interfaces and configured to execute one or more processes; and

a memory configured to store a process executable by the processor, the process when executed configured to:

apply labels to feature vectors of network characteristics associated with a plurality of Wi-Fi wireless access points in the network, wherein an applied label for a feature vector indicates whether a wireless access point associated with the feature vector experienced a threshold number of onboarding delays within a given time window, wherein an onboarding delay is determined by an amount of time exceeding a delay threshold, the amount of time indicative of when a client attempts to initially join the network or roams between wireless access points of the network;

train, based on the feature vectors and labels, a plurality of machine learning-based classifiers to predict onboarding delays;

use one or more of the trained plurality of classifiers to predict onboarding delays for a particular wireless access point;

calculate one or more classifier performance metrics for the one or more trained classifiers based on the predicted onboarding delays for the particular wireless access point and a history of onboarding delays; and

select a particular one of the trained classifiers to monitor the network characteristics associated with the particular wireless access point, based on the one or more calculated classifier performance metrics.

10. The apparatus as in claim 9 , wherein the process when executed is further configured to:

use the selected classifier to classify the network characteristics associated with the particular wireless access point, in order to predict onboarding delays for the particular wireless access point; and

initiate a change to the monitored network to mitigate the predicted onboarding delays for the particular wireless access point.

11. The apparatus as in claim 9 , wherein the one or more classifier performance metrics comprise at least one of: a classifier precision or a classifier recall.

12. The apparatus as in claim 9 , wherein the apparatus trains, based on the feature vectors and labels, a plurality of machine learning-based classifiers to predict onboarding delays by:

generating a plurality of subsets of the wireless access points in the network, each subset comprising wireless access points that have similar network characteristics; and

training, and for each subset of wireless access points, at least one machine learning-based classifier to predict onboarding delays using the feature vectors and labels associated with the wireless access points in the subset.

13. The apparatus as in claim 12 , wherein the apparatus generates the plurality of subsets of the wireless access points by:

using a machine learning-based clustering process on the network characteristics, to form the subsets of wireless access points.

14. The apparatus as in claim 9 , wherein the apparatus uses one or more of the trained plurality of classifiers to predict onboarding delays for the particular wireless access point by:

obtaining a history of network characteristics associated with the particular wireless access point from a defined time period; and

using one or more of the plurality of classifiers on the history of network characteristics associated with the particular wireless access point, to predict onboarding delays for the defined period of time.

15. The apparatus as in claim 14 , wherein the apparatus calculates the one or more classifier performance metrics for the one or more trained classifiers based on the predicted onboarding delays for the particular wireless access point by:

obtaining a history of onboarding delays from the defined time period; and

comparing the history of onboarding delays from the defined time period to the predicted onboarding delays for the defined time period.

16. The apparatus as in claim 9 , wherein the process when executed is further configured to:

determine that predicted onboarding delays for a second one of the wireless access points by one or more of the trained classifiers results in a performance metric for the one or more trained classifiers being below a threshold; and, in response

disable the plurality of classifiers from classifying network characteristics of the second wireless access point for a period of time.

17. A tangible, non-transitory, computer-readable medium storing program instructions that cause a network assurance service that monitors a wireless network to execute a process comprising:

applying, by the network assurance service, labels to feature vectors of network characteristics associated with a plurality of Wi-Fi wireless access points in the network, wherein an applied label for a feature vector indicates whether a wireless access point associated with the feature vector experienced a threshold number of onboarding delays within a given time window, wherein an onboarding delay is determined by an amount of time exceeding a delay threshold, the amount of time indicative of when a client attempts to initially join the network or roams between wireless access points of the network;

training, by the network assurance service and based on the feature vectors and labels, a plurality of machine learning-based classifiers to predict onboarding delays;

using, by the network assurance service, one or more of the trained plurality of classifiers to predict onboarding delays for a particular wireless access point;

calculating, by the network assurance service, one or more classifier performance metrics for the one or more trained classifiers based on the predicted onboarding delays for the particular wireless access point and a history of onboarding delays; and

selecting, by the network assurance service, a particular one of the trained classifiers to monitor the network characteristics associated with the particular wireless access point, based on the one or more calculated classifier performance metrics.