Machine learning techniques for generating impact measures using nonlinear causal inference and predictive risk scores

A system comprising memory and one or more processors communicatively coupled to the memory, the one or more processors configured to:

• generate a plurality of data object cohorts from a dataset, wherein the plurality of data object cohorts comprises:
• one or more positive cohorts with a first plurality of data objects associated with a predictive label;
• one or more negative cohorts with a second plurality of data objects unassociated with the predictive label;
• generate, using a positive causal inference model, a positive cohort impact measure for the one or more positive cohorts;
• generate, using a machine learning prediction model, a plurality of risk scores associated with the predictive label for the one or more negative cohorts;
• generate a positive risk-based cohort and a negative risk-based cohort from the one or more negative cohorts based on the plurality of risk scores;
• and
• generate an object impact measure for a particular data object of the positive risk-based cohort based on the plurality of risk scores, the positive cohort impact measure, and a negative cohort impact measure corresponding to the negative risk-based cohort.

One or more non-transitory, computer-readable storage media including instructions that, when executed by one or more processors, causes the one or more processors to:

• generate a plurality of data object cohorts from a dataset, wherein the plurality of data object cohorts comprises:
• one or more positive cohorts with a first plurality of data objects associated with a predictive label;
• one or more negative cohorts with a second plurality of data objects unassociated with the predictive label;
• generate, using a positive causal inference model, a positive cohort impact measure for the one or more positive cohorts;
• generate, using a machine learning prediction model, a plurality of risk scores associated with the predictive label for the one or more negative cohorts;
• generate a positive risk-based cohort and a negative risk-based cohort from the one or more negative cohorts based on the plurality of risk scores;
• and
• generate an object impact measure for a particular data object of the positive risk-based cohort based on the plurality of risk scores, the positive cohort impact measure, and a negative cohort impact measure corresponding to the negative risk-based cohort.