SYSTEMS AND METHODS FOR MERCHANT LEVEL FRAUD DETECTION BASED IN PART ON MERCHANT COHORT CLUSTERING
A method and apparatus for detecting fraudulent merchant activities at a commerce platform system are described. The method may include accessing signal data for a merchant system generated for one or more merchant system transactions processed by a commerce platform system, encoding the signal data into a plurality of sets of input signals, and inputting a first set of input signals into a first machine learning model to generate a fraud score from the first set of input signals. The method may also include performing clustering of the merchant system using a second set of input signals to cluster the merchant system with zero or more additional merchant systems in one of a plurality of different merchant system cohort clusters, and generating cluster fraud scores. In response to detecting that a cluster fraud score for a cluster into which the merchant system was assigned satisfies a cohort cluster fraud threshold, the method may include taking an action against each merchant system in the cluster.
1 . A method for detecting fraudulent merchant activities for a cluster of merchants at a commerce platform system, the method comprising:
accessing signal data for a merchant system generated for one or more merchant system transactions processed by a commerce platform system;
encoding the signal data into a plurality of sets of input signals;
inputting a first set of input signals into a first machine learning model, the first machine learning model trained to generate a fraud score from the first set of input signals;
performing clustering of the merchant system using a second set of input signals to cluster the merchant system with zero or more additional merchant systems in one of a plurality of different merchant system cohort clusters;
generating the fraud score using the first machine learning model, and generating one or more cluster assignments for the merchant system by performing the clustering;
generating, for each of the plurality of different merchant system cohort clusters, a corresponding cluster fraud score indicative of members of said each of the plurality of different merchant system cohort clusters being associated with fraudulent activities at the commerce platform system based at least in part on the fraud score; and
in response to detecting that a cluster fraud score for the one of the plurality of different merchant system cohort clusters into which the merchant system was clustered satisfies a cohort cluster fraud threshold, initiating one or more remediative actions against each merchant system that is a member of the one of the plurality of different merchant system cohort clusters.
2 . The method of claim 1 , wherein the first machine learning model comprises an ensemble of machine learning models, the ensemble of machine learning models comprising an XGBoost machine learning model and a Neural Network machine learning model, and wherein the clustering is performed using density-based spatial clustering or KMeans clustering.
3 . The method of claim 1 , wherein the first set of signal data comprises structure data input signals and unstructured data input signals, and wherein the second set of signal data comprises merchant system characteristics signal data indicative of merchant system activities at the commerce platform system.
4 . The method of claim 3 , wherein the merchant system characteristics signal data comprises a transaction volume over a period of time, an aggregate transaction amount of all commerce platform system over the period of time, an average transaction amount over the period of time, a merchant system location, a total number of fraud detections by the first machine learning model over a period of time, or a combination thereof.
5 . The method of claim 2 , further comprising:
generating a first cluster fraud score for a first merchant system cluster into which the merchant system was clustered by:
generating a first score indicative of a probability of fraud of the first merchant system cluster, the first score generated from the total number of merchant systems within the first cluster for which fraud has been detected during a transaction divided by a total number of merchant systems with the first cluster,
generating a second score indicative of an average of fraud scores associated with each merchant system within the first cluster,
generating a third score indicative of an average of maximum fraud scores determined for each merchant system within the first cluster,
determining the first cluster fraud score based on a combination of the first score, the second score, and the third score.
6 . The method of claim 5 , wherein determining the first cluster fraud score based on a combination of the first score, the second score, and the third score comprises determining the first cluster fraud score based on an average of the first score, the second score, and the third score.
7 . The method of claim 1 , wherein the clustering of the merchant system groups the merchant system with the zero or more additional merchant systems in one of a plurality of different merchant system cohort clusters based, at least in part, on an output generated by the first machine learning model for the merchant system.
8 . The method of claim 1 , wherein the clustering is performed on a periodic basis.
9 . The method of claim 1 , wherein initiating one or more remediative actions against each merchant system that is a member of the one of the plurality of different merchant system cohort clusters, comprises:
suspending accounts associated each merchant system in a fraudulent cluster for a period of time;
suspending accounts associated each merchant system in a fraudulent cluster until data indicative of non-fraudulent activity is provided by a merchant for which an account has been suspended;
deactivating accounts associated each merchant system in the fraudulent cluster; or
a combination thereof.
10 . A non-transitory computer readable storage medium storing instructions, which when executed by a computer processing system, causes the computer processing system to perform operations for detecting fraudulent merchant activities for a cluster of merchants at a commerce platform system, the operations comprising:
accessing signal data for a merchant system generated for one or more merchant system transactions processed by a commerce platform system;
encoding the signal data into a plurality of sets of input signals;
inputting a first set of input signals into a first machine learning model, the first machine learning model trained to generate a fraud score from the first set of input signals;
performing clustering of the merchant system using a second set of input signals to cluster the merchant system with zero or more additional merchant systems in one of a plurality of different merchant system cohort clusters;
generating the fraud score using the first machine learning model, and generating one or more cluster assignments for the merchant system by performing the clustering;
generating, for each of the plurality of different merchant system cohort clusters, a corresponding cluster fraud score indicative of members of said each of the plurality of different merchant system cohort clusters being associated with fraudulent activities at the commerce platform system based at least in part on the fraud score; and
in response to detecting that a cluster fraud score for the one of the plurality of different merchant system cohort clusters into which the merchant system was clustered satisfies a cohort cluster fraud threshold, initiating one or more remediative actions against each merchant system that is a member of the one of the plurality of different merchant system cohort clusters.
11 . The non-transitory computer readable storage medium of claim 10 , wherein the first machine learning model comprises an ensemble of machine learning models, the ensemble of machine learning models comprising an XGBoost machine learning model and a Neural Network machine learning model, and wherein the clustering is performed using density-based spatial clustering or KMeans clustering.
12 . The non-transitory computer readable storage medium of claim 10 , wherein the first set of signal data comprises structure data input signals and unstructured data input signals, and wherein the second set of signal data comprises merchant system characteristics signal data indicative of merchant system activities at the commerce platform system.
13 . The non-transitory computer readable storage medium of claim 12 , wherein the merchant system characteristics signal data comprises a transaction volume over a period of time, an aggregate transaction amount of all commerce platform system over the period of time, an average transaction amount over the period of time, a merchant system location, a total number of fraud detections by the first machine learning model over a period of time, or a combination thereof.
14 . The non-transitory computer readable storage medium of claim 12 , further comprising:
generating a first cluster fraud score for a first merchant system cluster into which the merchant system was clustered by:
generating a first score indicative of a probability of fraud of the first merchant system cluster, the first score generated from the total number of merchant systems within the first cluster for which fraud has been detected during a transaction divided by a total number of merchant systems with the first cluster,
generating a second score indicative of an average of fraud scores associated with each merchant system within the first cluster,
generating a third score indicative of an average of maximum fraud scores determined for each merchant system within the first cluster,
determining the first cluster fraud score based on a combination of the first score, the second score, and the third score.
15 . The non-transitory computer readable storage medium of claim 14 , wherein determining the first cluster fraud score based on a combination of the first score, the second score, and the third score comprises determining the first cluster fraud score based on an average of the first score, the second score, and the third score.
16 . The non-transitory computer readable storage medium of claim 10 , wherein the clustering of the merchant system groups the merchant system with the zero or more additional merchant systems in one of a plurality of different merchant system cohort clusters based, at least in part, on an output generated by the first machine learning model for the merchant system.
17 . The non-transitory computer readable storage medium of claim 10 , wherein the clustering is performed on a periodic basis.
18 . The non-transitory computer readable storage medium of claim 10 , wherein initiating one or more remediative actions against each merchant system that is a member of the one of the plurality of different merchant system cohort clusters, comprises:
suspending accounts associated each merchant system in a fraudulent cluster for a period of time;
suspending accounts associated each merchant system in a fraudulent cluster until data indicative of non-fraudulent activity is provided by a merchant for which an account has been suspended;
deactivating accounts associated each merchant system in the fraudulent cluster; or
a combination thereof.
19 . A server computer system for detecting fraudulent merchant activities for a cluster of merchants at a commerce platform system, the system comprising:
a memory; and
one or more processors coupled with the memory configured to perform operations, comprising:
accessing signal data for a merchant system generated for one or more merchant system transactions processed by a commerce platform system;
encoding the signal data into a plurality of sets of input signals;
inputting a first set of input signals into a first machine learning model, the first machine learning model trained to generate a fraud score from the first set of input signals;
performing clustering of the merchant system using a second set of input signals to cluster the merchant system with zero or more additional merchant systems in one of a plurality of different merchant system cohort clusters;
generating the fraud score using the first machine learning model, and generating one or more cluster assignments for the merchant system by performing the clustering;
generating, for each of the plurality of different merchant system cohort clusters, a corresponding cluster fraud score indicative of members of said each of the plurality of different merchant system cohort clusters being associated with fraudulent activities at the commerce platform system based at least in part on the fraud score; and
in response to detecting that a cluster fraud score for the one of the plurality of different merchant system cohort clusters into which the merchant system was clustered satisfies a cohort cluster fraud threshold, initiating one or more remediative actions against each merchant system that is a member of the one of the plurality of different merchant system cohort clusters.
20 . The server computer system of claim 19 , wherein the one or more processors are further configured to perform operations, comprising:
generating a first cluster fraud score for a first merchant system cluster into which the merchant system was clustered by:
generating a first score indicative of a probability of fraud of the first merchant system cluster, the first score generated from the total number of merchant systems within the first cluster for which fraud has been detected during a transaction divided by a total number of merchant systems with the first cluster,
generating a second score indicative of an average of fraud scores associated with each merchant system within the first cluster,
generating a third score indicative of an average of maximum fraud scores determined for each merchant system within the first cluster,
determining the first cluster fraud score based on a combination of the first score, the second score, and the third score.