Adaptive anomaly detector
A computer system is provided. The computer system includes a memory, a network interface, and a processor coupled to the memory and the network interface. The processor is configured to receive a response to a request to verify whether an ostensible client of a service is actually a client or a bot, the response including an indicator of whether the ostensible client is a client or a bot; receive information descriptive of interoperations between the ostensible client and the service that are indicative of whether the ostensible client is a client or a bot; and train a plurality of machine learning classifiers using the information and the indicator to generate a next generation of the plurality of machine learning classifiers.
1. A computer system comprising:
a memory;
a network interface; and
at least one processor coupled to the memory and the network interface and configured to
receive a response to a completely automated public Turing test to tell computers and humans apart (CAPTCHA) challenge to verify whether an ostensible client of a service is actually a client or a bot, the response including an indicator of whether the ostensible client is a client or a bot;
receive information descriptive of interoperations between the ostensible client and the service that are indicative of whether the ostensible client is a client or a bot;
train a plurality of machine learning classifiers using the information and the indicator to generate a next generation of the plurality of machine learning classifiers, wherein the plurality of machine learning classifiers includes a master classifier and one or more community classifiers;
calculate an accuracy of each machine learning classifier of the plurality of machine learning classifiers;
determine whether any community classifier of the next generation of machine learning classifiers has an accuracy that exceeds an accuracy of the master classifier by a first threshold amount; and
replace the master classifier with a particular community classifier of the one or more community classifiers to generate a new master classifier where the accuracy of the particular community classifier exceeds the accuracy of the master classifier by the first threshold amount.
2. The computer system of claim 1 , wherein the information descriptive of the interoperations includes one or more of source Internet Protocol (IP) address, destination IP address, source port, destination port, protocol, total packets exchanged, average inter arrival time of packets, and average time between mouse clicks.
3. The computer system of claim 1 , wherein the plurality of machine learning classifiers comprise a plurality of artificial neural networks.
4. The computer system of claim 1 , wherein the one or more community classifiers comprise a peer classifier, an alpha classifier, and an historical classifier.
5. The computer system of claim 1 , wherein the at least one processor is further configured to:
determine whether any particular machine learning classifier of the next generation of the plurality of machine learning classifiers has an accuracy that transgresses a threshold value based on the accuracy of an ancestor classifier of the particular machine learning classifier; and
replace the particular machine learning classifier with the ancestor classifier where the accuracy of the particular machine learning classifier transgresses the threshold value.
6. The computer system of claim 1 , wherein the at least one processor is further configured to:
identify a first community machine learning classifier of the next generation of the plurality of machine learning classifiers that has an accuracy that exceeds the accuracy of the master classifier by the first threshold amount; and
replace the master classifier with the first community machine learning classifier to generate a new master classifier.
7. The computer system of claim 6 , wherein the at least one processor is further configured to store the master classifier in a data store.
8. The computer system of claim 6 , wherein the plurality of machine learning classifiers are a first plurality of machine learning classifiers and the at least one processor is further configured to add, as a peer classifier, the new master classifier to a second plurality of machine learning classifiers comprising a second master classifier.
9. The computer system of claim 6 , wherein the memory includes an archive, and wherein the at least one processor is further configured to store in the archive the master classifier that was replaced by the first community classifier.
10. The computer system of claim 1 , wherein the at least one processor is further configured to:
determine that the accuracy of a particular community classifier of the next generation of machine learning classifiers has decreased by second threshold amount; and
replace the particular community classifier with a new community classifier.
11. The computer system of claim 1 , wherein the at least one processor is further configured to:
determine whether any particular machine learning classifier of the next generation of the plurality of machine learning classifiers has an accuracy that has decreased below a threshold value based on the accuracy of an ancestor classifier of the particular machine learning classifier; and
replace the particular machine learning classifier with a new machine learning classifier where the accuracy of the particular machine learning classifier has decreased below the threshold value.
12. A method of automatically training a plurality of machine learning classifiers to detect bots, the method comprising:
receiving a response to a completely automated public Turing test to tell computers and humans apart (CAPTCHA) challenge to verify whether an ostensible client of a service is actually a client or a bot, the response including an indicator of whether the ostensible client is a client or a bot;
receiving information descriptive of interoperations between the ostensible client and the service that are indicative of whether the ostensible client is a client or a bot;
training a plurality of machine learning classifiers using the information and the indicator to generate a next generation of the plurality of machine learning classifiers, wherein the plurality of machine learning classifiers includes a master classifier and one or more community classifiers;
calculating an accuracy of each machine learning classifier of the next generation of the plurality of machine learning classifiers;
determining whether any community classifier of the next generation of machine learning classifiers has an accuracy that exceeds an accuracy of the master classifier by a first threshold amount; and
replacing the master classifier with a particular community classifier of the one or more community classifiers to generate a new master classifier where the accuracy of the particular community classifier exceeds the accuracy of the master classifier by the first threshold amount.
13. The method of claim 12 , wherein receiving the information comprising receiving information including one or more of source Internet Protocol (IP) address, destination IP address, source port, destination port, protocol, total packets exchanged, average inter arrival time of packets, and average time between mouse clicks.
14. The method of claim 12 , further comprising:
determining whether any particular machine learning classifier of the next generation of the plurality of machine learning classifiers has an accuracy that transgresses a threshold value based on the accuracy of an ancestor classifier of the particular machine learning classifier; and
replacing the particular machine learning classifier with the ancestor classifier where the accuracy of the particular machine learning classifier transgresses the threshold value.
15. The method of claim 12 , further comprising:
identifying a first community classifier of the next generation of the plurality of machine learning classifiers that has an accuracy that exceeds the accuracy of the master classifier by the first threshold amount; and
replacing the master classifier with the first community classifier to generate a new master classifier.
16. The method of claim 15 , wherein the plurality of machine learning classifiers are a first plurality of machine learning classifiers and the method further comprises adding, as a peer classifier, the new master classifier to a second plurality of machine learning classifiers comprising a second master classifier.
17. The method of claim 15 , further comprising:
storing, in an archive, the master classifier that has been replaced by the first community classifier.
18. The method of claim 12 , further comprising:
determining whether the accuracy of any community classifier of the next generation of machine learning classifiers has decreased by a second threshold amount.
19. The method of claim 18 , further comprising:
determining that the accuracy of a first community classifier of the next generation of the plurality of machine learning classifiers has decreased by the second threshold amount; and
replacing the first community classifier with a new community classifier.
20. A non-transitory computer readable medium storing processor executable instructions to automatically train a plurality of machine learning classifiers, the instructions comprising instructions to:
receive a response to a completely automated public Turing test to tell computers and humans apart (CAPTCHA) challenge to verify whether an ostensible client of a service is actually a client or a bot, the response including an indicator of whether the ostensible client is a client or a bot;
receive information descriptive of interoperations between the ostensible client and the service that are indicative of whether the ostensible client is a client or a bot;
train a plurality of machine learning classifiers using the information and the indicator to generate a next generation of the plurality of machine learning classifiers, wherein the plurality of machine learning classifiers includes a master classifier and one or more community classifiers;
calculate an accuracy of each machine learning classifier of the next generation of the plurality of machine learning classifiers;
determine whether a particular community classifier of the next generation of the plurality of machine learning classifiers has an accuracy that transgresses a threshold value based on the accuracy of the master classifier; and
replace the master classifier with the particular community classifier to generate a new master classifier where the accuracy of the particular community classifier transgresses the threshold value.