Back to results
US 12,717,817 B2Granted
Systems and methods for time-based abnormality identification within uniform dataset
AssigneeOptum Services (Ireland) Limited
InventorsDamian Kelly, Julie Zhu
Life of this patent
Estimating the term…
Abstract
Various embodiments provide systems and methods for detecting data abnormalities within data sets relating to a particular agent and comprising both discrete and continuous data features by encoding the one or more discrete features and generating a sequential feature vector representative of both the encoded discrete features and the continuous features, reducing the dimensionality of the generated sequential feature vector to generate a reduced dimension behavioral vector, and comparing the reduced dimension behavioral vector against other reduced dimension behavioral vectors.
Claims22
A computer-implemented method comprising:
- receiving, by one or more processors, a set of agent-specific data, wherein first agent-specific data of the set of agent-specific data (i) is associated with a given agent of a plurality of agents associated with the set of agent-specific data and (ii) comprises (a) first continuous feature data comprising one or more continuous values for the given agent and (b) first discrete feature data for the given agent;
- encoding, by the one or more processors, a set of discrete feature data associated with the set of agent-specific data into a set of discrete feature vectors, wherein the set of discrete feature data comprises the first discrete feature data and encoding the first discrete feature data comprises encoding at least part of the first discrete feature data into one or more discrete feature vectors that are a subset of the set of discrete feature vectors;
- generating, by the one or more processors and using a sequential machine learning model, a set of sequential feature vectors based at least in part on the set of discrete feature vectors and continuous features data associated with the set of agent-specific data;
- reducing, by the one or more processors, dimensionality of the set of sequential feature vectors to generate a set of dimensionality-reduced behavior vectors associated with the plurality of agents;
- assigning, by the one or more processors, the set of dimensionality-reduced behavior vectors to a plurality of clusters based at least in part on determining distances between respective ones of the set of dimensionality-reduced behavior vectors;
- determining, by the one or more processors and from among the plurality of clusters, at least one non-fraudulent cluster that contains an agent, of the plurality of agents, associated with a dimensionality-reduced behavior vector of the set of dimensionality-reduced behavior vectors;
- determining, by the one or more processors and based at least in part on applying outlier detection to dimensionality-reduced behavior vector assigned to the agent contained in the at least one non-fraudulent cluster, that the dimensionality-reduced behavior vector should be associated with a fraudulent agent based at least in part on the dimensionality-reduced behavior vector deviating from a center of the at least one non-fraudulent cluster by a given distance;
- generating, by the one or more processors, instructions to display a representation of the agent contained in the at least one non-fraudulent cluster as fraudulent;
- and
- transmitting, by the one or more processors, the instructions to a user device.
7 more elements
A system comprising one or more processors and at least one memory storing processor-executable instructions that, when executed by the one or more processors, cause the one or more processors to perform operations comprising:
- receiving a set of agent-specific data, wherein first agent-specific data of the set of agent-specific data (i) is associated with a given agent of a plurality of agents associated with the set of agent-specific data and (ii) comprises (a) first continuous feature data comprising one or more continuous values for the given agent and (b) first discrete feature data for the given agent;
- encoding a set of discrete feature data associated with the set of agent-specific data into a set of discrete feature vectors, wherein the set of discrete feature data comprises the first discrete feature data and encoding the first discrete feature data comprises encoding at least part of the first discrete feature data into one or more discrete feature vectors that are a subset of the set of discrete feature vectors;
- generating, using a sequential machine learning model, a set of sequential feature vectors based at least in part on the set of discrete feature vectors and continuous features data associated with the set of agent-specific data;
- reducing dimensionality of the set of sequential feature vectors to generate a set of dimensionality-reduced behavior vectors associated with the plurality of agents;
- assigning one or more of the set of dimensionality-reduced behavior vectors to one or more of a plurality of clusters based at least in part on determining distances between respective ones of the set of dimensionality-reduced behavior vectors;
- determining, from among the plurality of clusters, at least one non-fraudulent cluster that contains an agent, of the plurality of agents, associated with a dimensionality-reduced behavior vector of the set of dimensionality-reduced behavior vectors;
- determining, based at least in part on applying outlier detection to the dimensionality-reduced behavior vector assigned to the agent contained in the at least one non-fraudulent cluster, that the dimensionality-reduced behavior vector should be associated with a fraudulent agent based at least in part on the dimensionality-reduced behavior vector deviating from a center of the at least one non-fraudulent cluster by a given distance;
- generating instructions to display a representation of the agent contained in the at least one non-fraudulent cluster as fraudulent;
- and
- transmitting the instructions to a user device.
7 more elements
One or more non-transitory computer readable storage media including instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:
- receiving a set of agent-specific data, wherein first agent-specific data of the set of agent-specific data (i) is associated with a given agent of a plurality of agents associated with the set of agent-specific data and (ii) comprises (a) first continuous feature data comprising one or more continuous values for the given agent and (b) first discrete feature data for the given agent;
- encoding a set of discrete feature data associated with the set of agent-specific data into a set of discrete feature vectors, wherein the set of discrete feature data comprises the first discrete feature data and encoding the first discrete feature data comprises encoding at least part of the first discrete feature data into one or more discrete feature vectors that are a subset of the set of discrete feature vectors;
- generating, using a sequential machine learning model, a set of sequential feature vectors based at least in part on the set of discrete feature vectors and continuous features data associated with the set of agent-specific data;
- reducing dimensionality of the set of sequential feature vectors to generate a set of dimensionality-reduced behavior vectors associated with the plurality of agents;
- assigning one or more of the set of dimensionality-reduced behavior vectors to one or more of a plurality of clusters based at least in part on determining distances between respective ones of the set of dimensionality-reduced behavior vectors;
- determining, from among the plurality of clusters, at least one non-fraudulent cluster that contains an agent, of the plurality of agents, associated with a dimensionality-reduced behavior vector of the set of dimensionality-reduced behavior vectors;
- determining, based at least in part on applying outlier detection to the dimensionality-reduced behavior vector assigned to the agent contained in the at least one non-fraudulent cluster, that the dimensionality-reduced behavior vector should be associated with a fraudulent agent based at least in part on the dimensionality-reduced behavior vector deviating from a center of the at least one non-fraudulent cluster by a given distance;
- generating instructions to display a representation of the agent contained in the at least one non-fraudulent cluster as fraudulent;
- and
- transmitting the instructions to a user device.
7 more elements
Dependencies are inferred from the claim text.