Triaging alerts using machine learning
In various embodiment, a process for triaging alerts using machine learning includes receiving data associated with transactions and using computer processor(s) to analyze the received data using rule(s) to automatically identify potentially suspicious activities. The process includes scoring each of the identified potentially suspicious activities using a machine learning model and based at least in part on analysis results of the rule(s) associated with the identified potentially suspicious activities, and triaging the identified potentially suspicious activities including by determining an action to take with respect to at least a portion of the identified potentially suspicious activities based at least in part on the scoring. In various embodiments, a process for training a machine learning model to triage alerts includes configuring the machine learning model, and receiving training data. The process includes using computer processor(s) and the received training data to train the machine learning model to score potentially suspicious activities.
1 . A method, comprising:
receiving data associated with transactions;
using one or more computer processors to analyze the received data using one or more rules to automatically identify potentially suspicious activities;
configuring a machine learning model;
receiving training data;
enriching the training data including by:
obtaining a graph snapshot of a prior time period;
updating the graph snapshot including by removing edges of a graph data representation of the transactions that no longer fall within at least one sliding window and adding at least one new edge corresponding to a current time period;
determining a graph feature for every node with an event in a target day for at least a portion of the received data; and
determining a node property feature including by:
performing a random walk that ends in response to at least one of:
reaching a known illicit node or an absence of available connections; and
calculating a metric associated with a comparison of successful random walks with respect to a total number of random walks, wherein a successful random walk ends in reaching the known illicit node;
using the one or more computer processors and the enriched training data to train the machine learning model to evaluate potentially suspicious activities;
scoring each of the identified potentially suspicious activities based at least in part on an output of the trained machine learning model and analysis results of the one or more rules associated with the identified potentially suspicious activities; and
triaging the identified potentially suspicious activities including by determining an action to take with respect to at least a portion of the identified potentially suspicious activities based at least in part on the scoring of each of the identified potentially suspicious activities.
2 . The method of claim 1 , further comprising extracting a first set of features and a second set of features associated with the received data, wherein:
the first set of features is used to analyze the received data using the one or more rules to automatically identify potentially suspicious activities, and
the second set of features is used to score each of the identified potentially suspicious activities using the machine learning model.
3 . The method of claim 1 , wherein triaging the identified potentially suspicious activities includes changing a priority of at least one of the identified potentially suspicious activities.
4 . The method of claim 1 , further comprising:
determining a scored list of the identified potentially suspicious activities; and
providing the scored list of the identified potentially suspicious activities.
5 . The method of claim 4 , further comprising:
receiving feedback associated with the scored list of the identified potentially suspicious activities; and
updating the machine learning model based on the received feedback.
6 . The method of claim 1 , further comprising:
determining an explanation for a decision associated with the triaging; and
providing the explanation.
7 . The method of claim 6 , wherein providing the explanation includes initially providing a second-level explanation that includes a grouping of features that belong to a same concept or semantic group.
8 . The method of claim 7 , wherein:
providing the explanation further includes providing a first-level explanation after providing the second-level explanation; and
the first-level explanation includes features and associated contributions used by the machine learning model to determine the decision associated with the automatic triaging.
9 . A system, comprising:
a processor configured to:
receive data associated with transactions;
use one or more computer processors to analyze the received data using one or more rules to automatically identify potentially suspicious activities;
configure a machine learning model;
receive training data;
enrich the training data including by:
obtaining a graph snapshot of a prior time period;
updating the graph snapshot including by removing edges of a graph data representation of the transactions that no longer fall within at least one sliding window and adding at least one new edge corresponding to a current time period;
determine a graph feature for every node with an event in a target day for at least a portion of the received data; and
determine a node property feature including by:
performing a random walk that ends in response to at least one of:
reaching a known illicit node or an absence of available connections; and
calculating a metric associated with a comparison of successful random walks with respect to a total number of random walks, wherein a successful random walk ends in reaching the known illicit node;
train, using the enriched training data, the machine learning model to evaluate potentially suspicious activities;
score each of the identified potentially suspicious activities based at least in part on an output of the trained machine learning model and analysis results of the one or more rules associated with the identified potentially suspicious activities; and
triage the identified potentially suspicious activities including by determining an action to take with respect to at least a portion of the identified potentially suspicious activities based at least in part on the scoring of each of the identified potentially suspicious activities; and
a memory coupled to the processor and configured to provide the processor with instructions.
10 . The system of claim 9 , wherein the processor is further configured to extract a first set of features and a second set of features associated with the received data, wherein:
the first set of features is used to analyze the received data using the one or more rules to automatically identify potentially suspicious activities, and
the second set of features is used to score each of the identified potentially suspicious activities using the machine learning model.
11 . A non-transitory computer readable medium storing computer instructions, when executed by a processor, causes the processor to:
receive data associated with transactions;
use one or more computer processors to analyze the received data using one or more rules to automatically identify potentially suspicious activities;
configure a machine learning model;
receive training data;
enrich the training data including by:
obtaining a graph snapshot of a prior time period;
updating the graph snapshot including by removing edges of a graph data representation that no longer fall within at least one sliding window and adding at least one new edge corresponding to a current time period;
determine a graph feature for every node with an event in a target day for at least a portion of the received data; and
determine a node property feature including by:
performing a random walk that ends in response to at least one of: reaching a known illicit node or an absence of available connections; and
calculating a metric associated with a comparison of successful random walks with respect to a total number of random walks, wherein a successful random walk ends in reaching the known illicit node;
train, using the enriched training data, the machine learning model to evaluate potentially suspicious activities;
score each of the identified potentially suspicious activities based at least in part on an output of the trained machine learning model and analysis results of the one or more rules associated with the identified potentially suspicious activities; and
triage the identified potentially suspicious activities including by determining an action to take with respect to at least a portion of the identified potentially suspicious activities based at least in part on the scoring of each of the identified potentially suspicious activities.
12 . The method of claim 1 , further comprising:
determining that at least one identified potentially suspicious activity is different from a previously-identified potentially suspicious activity;
form a set of potentially suspicious activities;
re-configure the machine learning model in response to a determination that a threshold amount of labeled data has been gathered for the formed set of potentially suspicious activities; and
in response to re-configuring the machine learning model, retraining the machine learning model to more accurately analyze potentially suspicious activities compared with a machine learning model that has not been re-trained.