Machine learning detection of fraudulent validation of financial institution credentials
A method for increasing computer network security. A first request to attach a first user account for a software program to an external secure account is received. The first request includes a first set of user credentials for accessing the external secure account and a first unverified email address, the external secure account controlled by a second external computer. After receiving the first request, features having metrics is received. The metrics describe at least creation of the first user account and include at least the first unverified email address. A machine learning model processing the features calculates a probability score that the first request is fraudulent. The probability score is compared to a threshold to form a comparison result. A security action is performed with respect to attaching the first user account to the external secure account in response to the comparison result.
1. A method for increasing security in a computer network, the method comprising:
receiving, in a software program executing on a first internal computer, a first request to attach a first user account for the software program to an external secure account, wherein the first request comprises a first set of user credentials for accessing the external secure account and a first unverified email address, the external secure account controlled by a second external computer;
receiving, after receiving the first request and at the first internal computer, a plurality of features comprising a plurality of metrics, the plurality of metrics describing at least creation of the first user account and including at least the first unverified email address;
calculating, by a machine learning model processing the plurality of features and executing on the first internal computer, a probability score that the first request is fraudulent;
comparing, by the first internal computer, the probability score to a first threshold to form a comparison result; and
performing, by the first internal computer, a security action with respect to attaching the first user account to the external secure account in response to the comparison result.
2. The method of claim 1 , further comprising:
receiving a second request to create a second user account;
repeating receiving the plurality of features;
calculating a second probability score that the second request is fraudulent, wherein calculating the second probability score takes into account a presence of the first request;
comparing the second probability score to a second threshold to form a second comparison result; and
performing, by the first internal computer, a second security action with respect to attaching the second user account to the external secure account in response to the second comparison result.
3. The method of claim 2 , further comprising:
responsive to receiving the second request to create the second user account, adjusting the second threshold to a lower number relative to the first threshold.
4. The method of claim 2 , wherein the second threshold is equal to or greater than the first threshold.
5. The method of claim 2 , further comprising:
responsive to receiving the second request to create the second user account, adjusting the second probability score upwardly relative to the probability score.
6. The method of claim 1 , wherein performing the security action comprises:
responsive to the probability score being less than the first threshold, transmitting instructions to create the first user account and transmitting instructions to attach the first user account to the external secure account.
7. The method of claim 1 , further comprising:
training the machine learning model prior to calculating using supervised machine learning.
8. The method of claim 7 , further comprising:
calculating, using a plurality of machine learning models, a corresponding plurality of probability scores, each of the plurality of machine learning models using the plurality of metrics, and wherein the plurality of metrics correspond to a known attempt to fraudulently attach the first user account to the external secure account;
determining which of the corresponding plurality of probability scores has a highest quality score;
selecting a particular probability score from the corresponding plurality of probability scores that has the highest quality score;
selecting a particular machine learning model from the plurality of machine learning models that corresponds to the highest quality score; and
using the particular machine learning model as the machine learning model.
9. A non-transitory computer readable storage medium storing instructions for increasing security in a computer network, the instructions, when executed by a computer processor, comprising:
program code for receiving a first request to attach a first user account for a software program to an external secure account, wherein the first request comprises a first set of user credentials for accessing the external secure account and a first unverified email address, the external secure account controlled by an external computer;
program code for receiving, after receiving the first request, a plurality of features comprising a plurality of metrics, the plurality of metrics describing at least creation of the first user account and including at least the first unverified email address;
program code for calculating, by a machine learning model processing the plurality of features, a probability score that the first request is fraudulent;
program code for comparing the probability score to a first threshold to form a comparison result; and
program code for performing a security action with respect to attaching the first user account to the external secure account in response to the comparison result.
10. The non-transitory computer readable storage medium of claim 9 , wherein the instructions further comprise:
program code for receiving a second request to create a second user account;
program code for repeating receiving the plurality of features;
program code for calculating a second probability score that the second request is fraudulent, wherein calculating the second probability score takes into account a presence of the first request;
program code for comparing the second probability score to a second threshold to form a second comparison result; and
program code for performing a second security action with respect to attaching the second user account to the external secure account in response to the second comparison result.
11. The non-transitory computer readable storage medium of claim 10 ,
wherein the instructions further comprise:
program code for, responsive to receiving the second request to create the second user account, adjusting the second threshold to a lower number relative to the first threshold.
12. The non-transitory computer readable storage medium of claim 10 , wherein the second threshold is equal to or greater than the first threshold.
13. The non-transitory computer readable storage medium of claim 10 , wherein the instructions further comprise:
program code for, responsive to receiving the second request to create the second user account, adjusting the second probability score upwardly relative to the probability score.
14. The non-transitory computer readable storage medium of claim 9 , wherein the instructions further comprise:
program code for, responsive to the probability score being less than the first threshold, transmitting instructions to create the first user account and transmitting instructions to attach the first user account to the external secure account.
15. The non-transitory computer readable storage medium of claim 9 , wherein the instructions further comprise:
program code for training the machine learning model prior to calculating using supervised machine learning.
16. The non-transitory computer readable storage medium of claim 15 , wherein the instructions further comprise:
program code for calculating, using a plurality of machine learning models, a corresponding plurality of probability scores, each of the plurality of machine learning models using the plurality of metrics, and wherein the plurality of metrics correspond to a known attempt to fraudulently attach the first user account to the external secure account;
program code for determining which of the corresponding plurality of probability scores has a highest quality score;
program code for selecting a particular probability score from the corresponding plurality of probability scores that has the highest quality score;
program code for selecting a particular machine learning model from the plurality of machine learning models that corresponds to the highest quality score; and
program code for using the particular machine learning model as the machine learning model.
17. A computer system comprising:
a processor; and
a data storage device storing security software executable by the processor, the security software comprising:
a data extractor configured to perform at least one of receiving and extracting a plurality of selected features from a plurality of data sources;
a machine learning engine comprising a machine learning model that takes as input the plurality of selected features and has as output a probability score representative of a probability that a given account attachment attempt by third-party software to attach an account to a sensitive data account is malicious;
an evaluation engine comprising a probability comparator that compares the probability score to a first threshold to form an evaluation result; and
a security engine configured to perform a security action based on the evaluation result.
18. The computer system of claim 17 , wherein the security engine is further configured to:
responsive to the probability score being less than the first threshold, transmit a signal to an account creator to create a user account and to transmit an account attachment request to the sensitive data account; and
responsive to the probability score being equal to or greater than the first threshold, to take the security action.
19. The computer system of claim 18 , wherein the security engine is further configured to perform at least one of:
instruct the account creator to deny attachment of the account to the sensitive data account;
deny transmission of a set of user credentials to the sensitive data account;
prevent further access to the third-party software;
monitor future use of the third-party software; and
attach the account to a false sensitive data account on a different server.
20. The computer system of claim 17 , wherein the machine learning engine further comprises a plurality of models and wherein the evaluation engine further comprises:
a model quality determinator configured to evaluate corresponding quality scores of the plurality of machine learning models; and
a model selector configured to select a particular machine learning model as having a highest quality score among the corresponding quality scores, the particular machine learning model being the machine learning model.