Systems and processes for bias removal in a predictive performance model

A system comprising:

• a memory device;
• and
• at least one hardware processor coupled to the memory device, wherein the at least one hardware processor is configured to at least:
• generate predictive model data comprising a plurality of compartmentalized variables;
• analyze the predictive model data to determine a plurality of bias values individually corresponding to a respective one of a plurality of protected data types;
• determine that the at least one of the plurality of bias values violates a bias threshold corresponding to the respective one of the plurality of protected data types;
• determine a respective contribution for each of the compartmentalized variables on the at least one of the plurality of bias values;
• identify a particular compartmentalized variable of the plurality of compartmentalized variables based on the respective contribution each of the plurality of compartmentalized variables;
• and
• generate second predictive model data comprising a plurality of second compartmentalized variables, the plurality of second compartmentalized variables excluding the particular compartmentalized variable.

A process comprising:

• generating predictive model data comprising a plurality of compartmentalized variables;
• analyzing the predictive model data to determine a plurality of bias values individually corresponding to a respective one of a plurality of protected data types;
• determining that the at least one of the plurality of bias values violates a bias threshold corresponding to the respective one of the plurality of protected data types;
• determining a respective contribution for each of the compartmentalized variables on the at least one of the plurality of bias values;
• identifying a particular compartmentalized variable of the plurality of compartmentalized variables based on the respective contribution each of the plurality of compartmentalized variables;
• and
• generating second predictive model data comprising a plurality of second compartmentalized variables, the plurality of second compartmentalized variables excluding the particular compartmentalized variable.