Search
Patexia Research
Patent No. US 10997511
Issue Date May 4, 2021
Claim this patent
PDF Unavailable

Patent 10997511 - Optimizing automated modeling algorithms for risk assessment and generation of explanatory data > Claims

  • 1. A system comprising: a processing device; anda memory device in which instructions executable by the processing device are stored for causing the processing device to:train a neural network model for computing a risk indicator from predictor variables, wherein the neural network model is a memory structure comprising nodes connected via one or more layers, wherein the risk indicator indicates a level of risk associated with an entity, wherein training the neural network model to generate a trained neural network model comprises: configuring the neural network model with common factors as input elements;retrieving predictor variables each corresponding to an action performed by an entity, wherein a portion of the predictor variables had been used to initialize the neural network model;obtaining common factors of the predictor variables, wherein each common factor is a single variable indicating a respective relationship among a respective subset of the predictor variables, and wherein non-monotonicity exists with respect to one or more of the common factors and the risk indicator,iteratively adjusting the memory structure defining the neural network according to monotonicity constraints and multicollinearity constraints to enforce (i) a monotonic relationship between each common factor and the risk indicator as determined by the neural network and (ii) a respective variance inflation factor for each common factor is below a threshold, wherein each variance inflation factor indicates multicollinearity among the common factors, adjusting the memory structure defining the neural network comprising one or more of rotating one or more of the common factors, removing one or more of the common factors, or setting one or more weights in the neural network to zero; andoutput, based on the trained neural network, explanatory data indicating relationships between (i) changes in the risk indicator and (ii) changes in at least some of the common factors.
    • 2. The system of claim 1, wherein the processing device is further configured to: determine specific factors by performing a factor analysis on the predictor variables, wherein each specific factor includes unique information associated with a respective predictor variable, wherein the unique information is not included in common factors corresponding to the respective predictor variable,iteratively adjust the neural network to enforce a respective additional variance inflation factor for each specific factor to be within a specific factor threshold.
      • 3. The system of claim 2, wherein the processing device is further configured to generate the explanatory data by performing operations comprising: identifying a risk-assessment function having (i) the common factors and the specific factors as inputs and (ii) the risk indicator as an output;computing risk-indicator decreases for the common factors, wherein the processing device is configured to compute each risk-indicator decrease by at least:determining a respective maximum value of the risk indicator using (i) constant values for the specific factors and (ii) a value of a respective common factor that maximizes the risk-assessment function, determining a respective decreased value of the risk indicator using (i) the constant values for the specific factors and (ii) a value of the respective common factor for the entity, anddetermining the risk-indicator decrease from a difference between the respective maximum value and the respective decreased value; andgenerating the explanatory data from a subset of the risk-indicator decreases having the largest values.
      • 4. The system of claim 2, wherein the processing device is further configured to generate the explanatory data by performing operations comprising: identifying a risk-assessment function having (i) the common factors and the specific factors as inputs and (ii) the risk indicator as an output;computing risk-indicator decreases, wherein the risk-indicator decreases comprise (i) first risk-indicator decreases for the common factors and (ii) second risk-indicator decreases for the specific factors; andgenerating the explanatory data from a subset of the risk-indicator decreases having the largest values,wherein the processing device is configured to compute each first risk-indicator decrease by at least: determining a respective maximum value of the risk indicator using (i) a value of a respective common factor that maximizes the risk-assessment function and (ii) constant values for the specific factors,determining a respective decreased value of the risk indicator using (i) a value of the respective common factor for the entity and (ii) the constant values for the specific factors, anddetermining the first risk-indicator decrease from a difference between the respective maximum value and the respective decreased value,wherein the processing device is configured to compute each second risk-indicator decrease by at least: determining a respective maximum value of the risk indicator using (i) a value of a respective specific factor that maximizes the risk-assessment function and (ii) constant values for the common factors,determining a respective decreased value of the risk indicator using (i) a different value of the respective specific factor and (ii) the constant values for the common factors, anddetermining the second risk-indicator decrease from a difference between the respective maximum value and the respective decreased value;wherein the processing device is configured to use a subset of the risk-indicator decreases having the largest values to generate the explanatory data.
        • 5. The system of claim 4, wherein the processing device is configured to adjust the neural network by eliminating connections in the neural network involving at least one of: relationships not in accordance with the expected trend between specific factors and the risk indicator as determined by the neural network; orexcessive variance inflation factors that exceed the additional threshold.
    • 6. The system of claim 1, wherein the processing device is further configured to generate the explanatory data by performing operations comprising: identifying specific factors generated by performing the factor analysis on the predictor variables, wherein each specific factor includes unique information associated with a respective predictor variable, wherein the unique information is not captured by common factors corresponding to the respective predictor variable;identifying a risk-assessment function having (i) the common factors and the specific factors as inputs and (ii) the risk indicator as an output;assigning zero-values to the specific factors;computing risk-indicator decreases for the common factors, wherein the processing device is configured to compute each risk-indicator decrease by at least: determining a respective maximum value of the risk indicator using (i) a value of a respective common factor that maximizes the risk-assessment function and (ii) the zero-values for the specific factors,determining a respective decreased value of the risk indicator using (i) a value of the respective common factor for the entity and (ii) the zero-values for the specific factors, anddetermining the risk-indicator decrease from a difference between the respective maximum value and the respective decreased value; andgenerating the explanatory data from a subset of the risk-indicator decreases having the largest values.
    • 7. The system of claim 1, wherein obtaining the common factors of the predictor variables comprises performing a factor analysis on the predictor variables to identify the common factors of the predictor variables.
    • 8. The system of claim 1, wherein the processing device is configured to adjust the neural network by adjusting at least one of: a number of nodes in a hidden layer of the neural network,a connection in the neural network,the predictor variables, ora number of layers in the hidden neural network.
    • 9. The system of claim 1, wherein the processing device is configured to identify the predictor variables by performing operations comprising: identifying a set of candidate predictor variables;identifying, for each of the candidate predictor variables, a respective bivariate relationship between the candidate predictor variable and the outcome; andtransforming, based on the identified bivariate relationships, the set of candidate predictor variables into the predictor variables.
  • 10. A non-transitory computer-readable medium having program code that is executable by a processing device to perform operations, the operations comprising: train a neural network model for computing a risk indicator from predictor variables, wherein the neural network model is a memory structure comprising nodes connected via one or more layers, wherein the risk indicator indicates a level of risk associated with an entity, wherein training the neural network model to generate a trained neural network model comprises: configuring the neural network model with common factors as input elements;retrieving predictor variables each corresponding to an action performed by an entity, wherein a portion of the predictor variables had been used to initialize the neural network model;obtaining common factors of the predictor variables, wherein each common factor is a single variable indicating a respective relationship among a respective subset of the predictor variables, and wherein non-monotonicity exists with respect to one or more of the common factors and the risk indicator;iteratively adjusting the memory structure defining the neural network according to monotonicity constraints and multicollinearity constraints to enforce (i) a monotonic relationship between each common factor and the risk indicator as determined by the neural network and (ii) a respective variance inflation factor for each common factor is below a threshold, wherein each variance inflation factor indicates multicollinearity among the common factors, adjusting the memory structure defining the neural network comprising one or more of rotating one or more of the common factors, removing one or more of the common factors, or setting one or more weights in the neural network to zero; andoutputting based on the trained neural network, explanatory data indicating relationships between (i) changes in the risk indicator and (ii) changes in at least some of the common factors.
    • 11. The non-transitory computer-readable medium of claim 10, wherein the operations further comprising: determining specific factors by performing the factor analysis on the predictor variables, wherein each specific factor includes unique information associated with a respective predictor variable, wherein the unique information is not included in common factors corresponding to the respective predictor variable; anditeratively adjusting the neural network to enforce a respective additional variance inflation factor for each specific factor to be within a specific factor threshold.
      • 12. The non-transitory computer-readable medium of claim 11, wherein generating the explanatory data comprises: identifying a risk-assessment function having (i) the common factors and the specific factors as inputs and (ii) the risk indicator as an output;computing risk-indicator decreases for the common factors, wherein computing each risk-indicator decrease comprises: determining a respective maximum value of the risk indicator using (i) constant values for the specific factors and (ii) a value of a respective common factor that maximizes the risk-assessment function,determining a respective decreased value of the risk indicator using (i) the constant values for the specific factors and (ii) a value of the respective common factor for the entity, anddetermining the risk-indicator decrease from a difference between the respective maximum value and the respective decreased value; andgenerating the explanatory data from a subset of the risk-indicator decreases having the largest values.
      • 13. The non-transitory computer-readable medium of claim 11, wherein generating the explanatory data comprises: identifying a risk-assessment function having (i) the common factors and the specific factors as inputs and (ii) the risk indicator as an output;computing risk-indicator decreases, wherein the risk-indicator decreases comprise (i) first risk-indicator decreases for the common factors and (ii) second risk-indicator decreases for the specific factors; andgenerating the explanatory data from a subset of the risk-indicator decreases having the largest values,wherein computing each first risk-indicator decrease comprises: determining a respective maximum value of the risk indicator using (i) a value of a respective common factor that maximizes the risk-assessment function and (ii) constant values for the specific factors,determining a respective decreased value of the risk indicator using (i) a value of the respective common factor for the entity and (ii) the constant values for the specific factors, anddetermining the first risk-indicator decrease from a difference between the respective maximum value and the respective decreased value,wherein computing each second risk-indicator decrease comprises: determining a respective maximum value of the risk indicator using (i) a value of a respective specific factor that maximizes the risk-assessment function and (ii) constant values for the common factors,determining a respective decreased value of the risk indicator using (i) a different value of the respective specific factor and (ii) the constant values for the common factors, anddetermining the second risk-indicator decrease from a difference between the respective maximum value and the respective decreased value; andwherein a subset of the risk-indicator decreases having the largest values is used to generate the explanatory data.
        • 14. The non-transitory computer-readable medium of claim 13, wherein adjusting the neural network comprises eliminating connections in the neural network involving at least one of:relationships not in accordance with the expected trend between specific factors and the risk indicator as determined by the neural network; orexcessive variance inflation factors that exceed the additional threshold.
    • 15. The non-transitory computer-readable medium of claim 10, wherein generating the explanatory data comprises: identifying specific factors generated by performing the factor analysis on the predictor variables, wherein each specific factor includes unique information associated with a respective predictor variable, wherein the unique information is not captured by common factors corresponding to the respective predictor variable;identifying a risk-assessment function having (i) the common factors and the specific factors as inputs and (ii) the risk indicator as an output;assigning zero-values to the specific factors; andcomputing risk-indicator decreases for the common factors, wherein computing each risk-indicator decrease comprises: determining a respective maximum value of the risk indicator using (i) a value of a respective common factor that maximizes the risk-assessment function and (ii) the zero-values for the specific factors,determining a respective decreased value of the risk indicator using (i) a value of the respective common factor for the entity and (ii) the zero-values for the specific factors, anddetermining the risk-indicator decrease from a difference between the respective maximum value and the respective decreased value; andgenerating the explanatory data from a subset of the risk-indicator decreases having the largest values.
    • 16. The non-transitory computer-readable medium of claim 10, wherein generating the explanatory data comprises: identifying specific factors generated by performing the factor analysis on the predictor variables, wherein each specific factor includes unique information associated with a respective predictor variable, wherein the unique information is not captured by common factors corresponding to the respective predictor variable;identifying a risk-assessment function having (i) the common factors and the specific factors as inputs and (ii) the risk indicator as an output;assigning zero-values to a first subset of the specific factors having relationships with respect to the risk indicator not in accordance with the expected trend;computing risk-indicator decreases, wherein the risk-indicator decreases comprise (i) first risk-indicator decreases for the common factors and (ii) second risk-indicator decreases for a second subset of the specific factors; andgenerating the explanatory data from a subset of the risk-indicator decreases having the largest values,wherein computing each first risk-indicator decrease comprises: determining a respective maximum value of the risk indicator using (i) a value of a respective common factor that maximizes the risk-assessment function, (ii) the zero-values for the first subset of the specific factors, and (iii) constant values for the second subset of the specific factors,determining a respective decreased value of the risk indicator using (i) a value of the respective common factor for the entity, (ii) the zero-values for the first subset of the specific factors, and (iii) constant values for the second subset of the specific factors, anddetermining the first risk-indicator decrease from a difference between the respective maximum value and the respective decreased value,wherein computing each second risk-indicator decrease comprises: determining a respective maximum value of the risk indicator using (i) constant values for the common factors, (ii) zero values for the first subset of the specific factors, and (iii) a value of a respective specific factor, from the second subset of the specific factors, that maximizes the risk-assessment function, determining a respective decreased value of the risk indicator using (i) the constant values for the common factors, (ii) zero values for the first subset of the specific factors, and (iii) a value of the respective specific factor for the entity that is selected from the second subset of the specific factors, anddetermining the second risk-indicator decrease from a difference between the respective maximum value and the respective decreased value;wherein a subset of the risk-indicator decreases having the largest values is used to generate the explanatory data.
    • 17. The non-transitory computer-readable medium of claim 10, wherein adjusting the neural network comprises adjusting at least one of: a number of nodes in a hidden layer of the neural network,a connection in the neural network,the predictor variables, ora number of layers in the hidden neural network.
    • 18. The non-transitory computer-readable medium of claim 10, wherein the operations further comprise identifying the predictor variables by at least: identifying a set of candidate predictor variables;identifying, for each of the candidate predictor variables, a respective bivariate relationship between the candidate predictor variable and the outcome; andtransforming, based on the identified bivariate relationships, the set of candidate predictor variables into the predictor variables.
  • 19. A method comprising: train a neural network model for computing a risk indicator from predictor variables, wherein the neural network model is a memory structure comprising nodes connected via one or more layers, wherein the risk indicator indicates a level of risk associated with an entity, wherein training the neural network model to generate a trained neural network model comprises: configuring the neural network model with common factors as input elements;retrieving predictor variables each corresponding to an action performed by an entity, wherein a portion of the predictor variables had been used to initialize the neural network model;obtaining common factors of the predictor variables, wherein each common factor is a single variable indicating a respective relationship among a respective subset of the predictor variables, and wherein non-monotonicity exists with respect to one or more of the common factors and the risk indicator,iteratively adjusting the memory structure defining the neural network according to monotonicity constraints and multicollinearity constraints to enforce (i) a monotonic relationship between each common factor and the risk indicator as determined by the neural network and (ii) a respective variance inflation factor for each common factor is below a threshold, wherein each variance inflation factor indicates multicollinearity among the common factors, adjusting the memory structure defining the neural network comprising one or more of rotating one or more of the common factors, removing one or more of the common factors, or setting one or more weights in the neural network to zero; andoutputting, based on the trained neural network, explanatory data indicating relationships between (i) changes in the risk indicator and (ii) changes in at least some of the common factors.
    • 20. The method of claim 19, further comprising: determining specific factors by performing a factor analysis on the predictor variables, wherein each specific factor includes unique information associated with a respective predictor variable, wherein the unique information is not included in common factors corresponding to the respective predictor variable,iteratively adjusting the neural network to enforce a respective additional variance inflation factor for each specific factor to be within a specific factor threshold,wherein adjusting the neural network comprises eliminating connections in the neural network involving at least one of: relationships not in accordance with the expected trend between specific factors and the risk indicator as determined by the neural network; orexcessive variance inflation factors that exceed the additional threshold.
Menu