Classification and forwarding of network traffic flows

1. A non-transitory machine-readable medium comprising instructions that, when executed, cause a processing resource of a computing device to:

receive a traffic flow over a network;

classify, using a first machine learning (ML) model, the traffic flow into a first traffic class from a plurality of traffic classes based on flow characteristics of the traffic flow;

determine, from the plurality of traffic classes, a second traffic class based on computing a deviation of the flow characteristics of the traffic flow from average flow characteristics of each of the plurality of traffic classes and identifying a traffic class associated with the average flow characteristics with a least deviation;

update a quality metric for the first ML model based on whether the first traffic class matches the second traffic class, wherein the quality metric is a ratio of a frequency of a match between the first traffic class and the second traffic class (f match ) to the sum of frequencies of the match between the first traffic class and the second traffic class match, (f match ) and a mismatch between the first traffic class and the second traffic class (f mismatch ); and

selectively forward the traffic flow over the network based on the quality metric being within a range of threshold values indicating that the first traffic class matches the second traffic class, wherein selectively forwarding includes permitting, denying, prioritizing, or rerouting the traffic flow.

2. The non-transitory machine-readable medium of claim 1 , further comprising instructions to:

in response to a determination that the quality metric is not in a range of threshold values:

transmit, to an administration device, a notification indicative of a performance of the first ML model.

3. The non-transitory machine-readable medium of claim 1 , wherein the instructions to selectively forward the traffic flow further comprising instructions to:

in response to a determination that the quality metric is in a range of threshold values and that the first traffic class matches the second traffic class:

forward the traffic flow to a destination device corresponding to a destination address indicated in a metadata of the traffic flow.

4. The non-transitory machine-readable medium of claim 1 , wherein the instructions to selectively forward the traffic flow further comprising instructions to:

in response to a determination that the quality metric is in a range of threshold values and that the first traffic class matches the second traffic class:

determine, based on a metadata of the traffic flow, a source device of the traffic flow classified into the first traffic class, wherein the first traffic class is associated with non-malicious traffic flows; and

prioritize the traffic flow received from the source device.

5. The non-transitory machine-readable medium of claim 1 , wherein the instructions to selectively forward the traffic flow further comprising instructions to:

in response to a determination that the quality metric is in a range of threshold values and that the first traffic class matches the second traffic class:

determine, based on a metadata of the traffic flow, a source device of the traffic flow classified into the first traffic class, wherein the first traffic class is associated with resource intensive traffic flows; and

reroute the traffic flow from the source device to a predetermined destination device with higher resource availability.

6. The non-transitory machine-readable medium of claim 1 , wherein the instructions to update the quality metric further comprising instructions to:

recompute the ratio of the frequency of the match between the first traffic class and the second traffic class (f match ) to the sum of frequencies of the match between the first traffic class and the second traffic class (f match ) and the mismatch between the first traffic class and the second traffic class (f mismatch ).

7. The non-transitory machine-readable medium of claim 1 , further comprising instructions to determine the average flow characteristics of each of the plurality of traffic classes using a second ML model.

8. A device comprising:

a processing resource; and

a non-transitory machine-readable medium storing instructions that, when executed by the processing resource, cause the processing resource to:

receive a traffic flow over a network;

classify, using a first machine learning (ML) model, the traffic flow into a first traffic class from a plurality of traffic classes based on flow characteristics of the traffic flow;

determine, from the plurality of traffic classes, a second traffic class based on computing a deviation of the flow characteristics of the traffic flow from average flow characteristics of each of the plurality of traffic classes and identifying a traffic class associated with the average flow characteristics with a least deviation;

update a quality metric for the first ML model based on whether the first traffic class matches the second traffic class, wherein the quality metric is a ratio of a frequency of a match between the first traffic class and the second traffic class (f match ) to the sum of frequencies of the match between the first traffic class and the second traffic class match, (f match ) and a mismatch between the first traffic class and the second traffic class (f mismatch ); and

perform a traffic filtering action on the traffic flow based on an access control list, wherein the access control list indicates rules corresponding to the quality metric being within a range of threshold values indicating a match between the first traffic class and the second traffic class and a flag indicating the match, and wherein the traffic filtering action includes permitting, denying, prioritizing, or rerouting the traffic flow.

9. The device of claim 8 , wherein the processing resource executes one or more of the instructions to: create the rules to perform the traffic filtering action on the traffic flow based on the quality metric.

10. The device of claim 9 , wherein the access control list includes a plurality of entries, each entry specifying the rules and one or more of the quality metric, the flag, the first traffic class, a source address, or a destination address.

11. The device of claim 9 , wherein the processing resource executes one or more of the instructions to:

in response to a determination that the quality metric is in a range of threshold values and that the flag indicates that the first traffic class matches the second traffic class:

perform the traffic filtering action, wherein the traffic filtering action includes permitting the traffic flow to a destination device corresponding to a destination address indicated in a metadata of the traffic flow.

12. The device of claim 9 , wherein the processing resource executes one or more of the instructions to:

in response to a determination that the quality metric is not in a range of threshold values:

perform the traffic filtering action, wherein the traffic filtering action includes notifying an administration device of a performance of the first ML model.

13. The device of claim 9 , wherein the processing resource executes one or more of the instructions to:

in response to a determination that the quality metric is in a range of threshold values and that the flag indicates that the first traffic class matches the second traffic class, and wherein the first traffic class is associated with malicious traffic flows:

perform the traffic filtering action, wherein the traffic filtering action includes denying the traffic flow.

14. The device of claim 8 , wherein the flow characteristics include one or more of a packet length data, packet arrival time, source Internet Protocol (IP) address, destination IP address, a source port, a destination port, an IP protocol type, type of service, quality of service (QoS), timestamps, a number of packets, or a number of bytes associated with the traffic flow.

15. A method comprising:

receiving a traffic flow over a network;

classifying, using a first machine learning (ML) model, the traffic flow into a first traffic class from a plurality of traffic classes based on flow characteristics of the traffic flow;

determining, from the plurality of traffic classes, a second traffic class based on computing a deviation of the flow characteristics of the traffic flow from average flow characteristics of each of the plurality of traffic classes and identifying a traffic class associated with the average flow characteristics with a least deviation;

updating a quality metric for the first ML model based on whether the first traffic class matches the second traffic class, wherein the quality metric is a ratio of a frequency of a match between the first traffic class and the second traffic class (f match ) to the sum of frequencies of the match between the first traffic class and the second traffic class (f match ) and a mismatch between the first traffic class and the second traffic class (f mismatch ); and

selectively forwarding the traffic flow over the network based on the quality metric being within a range of threshold values indicating that the first traffic class matches the second traffic class, wherein selectively forwarding includes permitting, denying, prioritizing, or rerouting the traffic flow.

16. The method of claim 15 , further comprising determining the average flow characteristics of each of the plurality of traffic classes using a second ML model.

17. The method of claim 15 , further comprising:

in response to determining that the quality metric is in a range of threshold values and that the first traffic class does not match with the second traffic class:

providing, to an administration device, a notification indicative of a potential misclassification of the traffic flow by the first ML model.

18. The method of claim 15 , wherein updating the quality metric includes recomputing the ratio of the frequency of the match between the first traffic class and the second traffic class match, to (f match ) the sum of frequencies of the match between the first traffic class and the second traffic class (f match ) and the mismatch between the first traffic class and the second traffic class (f mismatch ).