Automated rule generation system and methods

1. A system adapted to automatically generate and validate rules for monitoring suspicious activity, the system comprising:

a processor and a computer readable medium operably coupled thereto, the computer readable medium comprising a plurality of instructions stored in association therewith that are accessible to, and executable by, the processor, to perform operations which comprise:

for a first period of time:

collecting a first group of transactions issued by an issuer;

automatically identifying and storing a plurality of key indicators of the first group of transactions;

automatically storing pre-identified fraudulent transactions of the first group of transactions;

based on the plurality of key indicators and the pre-identified fraudulent transactions, creating a training dataset;

using the training dataset:

training a custom decision tree machine learning algorithm; and

generating a feature depth map comprising a user-definable number of layers;

with the custom decision tree machine learning algorithm, generating a decision tree incorporating logical predicates including at least one key indicator of the plurality of key indicators by:

for each layer of the feature depth map:

if the layer contains a feature, fetching the feature from the feature depth map;

if the layer does not contain a feature, fetching a random feature from the training dataset;

selecting a split for the fetched feature based on a dispersion score; and

using the split, creating left and right child nodes for that layer,

wherein no two of the logical predicates within the decision tree are the same;

based on the decision tree, generating a plurality of rules, wherein each rule of the plurality of rules incorporates only one logical predicate from each layer of the decision tree; and

for a second period of time, in real time:

collecting a second group of transactions issued by an issuer;

generating a quality metric for each respective rule of the plurality of rules, by automatically testing the plurality of rules against the second group of transactions;

identifying a subset of rules of the plurality of rules for which the respective quality metric exceeds a threshold value; and

displaying the subset of rules on a display,

wherein the displayed subset of rules comprises fewer, rules with fewer logical predicates than a set of rules generated from the first group of transactions wherein each rule of the plurality of rules incorporates a plurality of logical predicates from each layer of the decision tree.

2. The system of claim 1 , wherein the plurality of key indicators includes at least one of a transaction value for at least one transaction of the first group of transactions, a volume of transactions of the issuer, a credit limit of the issuer, a risk category of the issuer, or a net worth of the issuer, or a ratio of any two of the foregoing.

3. The system of claim 2 , wherein the plurality of key indicators includes at least one of a monthly pattern or a weekly pattern of the at least one key indicator of the plurality of key indicators.

4. The system of claim 1 , wherein the operations further comprise, for each respective rule of the plurality of rules, if a first logical predicate of the respective rule is logically redundant with a second logical predicate of the respective rule, deleting the first logical predicate of the respective rule.

5. The system of claim 1 , wherein the quality metric comprises at least one of a number or fraction of true positives, a number or fraction of false positives, a precision value, a recall value, an F1 value, or an FBeta value.

6. The system of claim 1 , wherein the quality metric comprises a number of logical predicates within the respective rule.

7. The system of claim 1 , wherein the operations further comprise:

with the learning algorithm, generating a plurality of decision trees, wherein each decision tree of the plurality of decision trees incorporates logical predicates including at one of the plurality of key indicators, wherein no two of the logical predicates within any tree of the plurality of decision trees are the same; and

based on each respective decision tree of the plurality of decision trees, generating a plurality of additional rules of the plurality of rules, wherein each additional rule of the plurality of rules incorporates only one logical predicate from each layer of the respective decision tree of the plurality of decision trees.

8. The system of claim 1 , wherein the operations further comprise:

generating a respective plurality of quality metrics for each respective rule of the plurality of rules, by automatically testing the plurality of rules against the second group of transactions; and

identifying the subset of rules of the plurality of rules based on whether any respective quality metric of the respective plurality of quality metrics exceeds a respective threshold for that respective quality metric.

9. The system of claim 1 , wherein the learning algorithm is a rule-based machine learning algorithm.

10. The system of claim 9 , wherein the rule-based machine learning algorithm is a learning classifier system, association rule learning system, or artificial immune system.

11. A computer-implemented method adapted to automatically generate and validate rules for monitoring suspicious activity, the method comprising:

for a first period of time:

collecting a first group of transactions issued by an issuer;

automatically identifying and storing a plurality of key indicators of the first group of transactions;

automatically storing pre-identified fraudulent transactions of the first group of transactions;

based on the plurality of key indicators and the pre-identified fraudulent transactions, creating a training dataset;

using the training dataset:

training a custom decision tree machine learning algorithm; and

generating a feature depth map comprising a user-definable number of layers;

with the custom decision tree machine learning algorithm, generating a decision tree incorporating logical predicates including at least one key indicator of the plurality of key indicators by:

for each layer of the feature depth map:

if the layer contains a feature, fetching the feature from the feature depth map;

if the layer does not contain a feature, fetching a random feature from the training dataset;

selecting a split for the fetched feature based on a dispersion score; and

using the split, creating left and right child nodes for that layer,

wherein no two of the logical predicates within the decision tree are the same;

based on the decision tree, generating a plurality of rules, wherein each rule of the plurality of rules incorporates only one logical predicate from each layer of the decision tree; and

for a second period of time, in real time:

collecting a second group of transactions issued by an issuer;

generating a quality metric for each respective rule of the plurality of rules, by automatically testing the plurality of rules against the second group of transactions;

identifying a subset of rules of the plurality of rules for which the respective quality metric exceeds a threshold value; and

displaying the subset of rules on a display,

wherein the displayed subset of rules comprises fewer, rules with fewer logical predicates than a set of rules generated from the first group of transactions wherein each rule of the plurality of rules incorporates a plurality of logical predicates from each layer of the decision tree.

12. The computer-implemented method of claim 11 , wherein the plurality of key indicators includes at least one of a transaction value for at least one transaction of the first group of transactions, a volume of transactions of the issuer, a credit limit of the issuer, a risk category of the issuer, or a net worth of the issuer, or a ratio of any two of the foregoing.

13. The computer-implemented method of claim 12 , wherein the plurality of key indicators includes at least one of a monthly pattern or a weekly pattern of the at least one key indicator of the plurality of key indicators.

14. The computer-implemented method of claim 11 , further comprising, for each respective rule of the plurality of rules, if a first logical predicate of the respective rule is logically redundant with a second logical predicate of the respective rule, deleting the first logical predicate of the respective rule.

15. The computer-implemented method of claim 11 , wherein the quality metric comprises at least one of a number or fraction of true positives, a number or fraction of false positives, a precision value, a recall value, an F1 value, or an FBeta value.

16. The computer-implemented method of claim 11 , wherein the quality metric comprises a number of logical predicates within the respective rule.

17. The computer-implemented method of claim 11 , further comprising:

with the learning algorithm, generating a plurality of decision trees, wherein each decision tree of the plurality of decision trees incorporates logical predicates including at least one of the plurality of key indicators, wherein no two of the logical predicates within any tree of the plurality of decision trees are the same; and

based on each respective decision tree of the plurality of decision trees, generating a plurality of additional rules of the plurality of rules, wherein each additional rule of the plurality of rules incorporates only one logical predicate from each layer of the respective decision tree of the plurality of decision trees.

18. The computer-implemented method of claim 11 , further comprising:

generating a respective plurality of quality metrics for each respective rule of the plurality of rules, by automatically testing the plurality of rules against the second group of transactions; and

identifying the subset of rules of the plurality of rules based on whether any respective quality metric of the respective plurality of quality metrics exceeds a respective threshold for that respective quality metric.

19. The computer-implemented method of claim 11 , wherein the learning algorithm is a rule-based machine learning algorithm.

20. The computer-implemented method of claim 19 , wherein the rule-based machine learning algorithm is a learning classifier system, association rule learning system, or artificial immune system.