Systems and methods for improving the interpretability and transparency of machine learning models

A method for determining, at a record level, impacts of particular input variables on a machine learning model, the method comprising:

• receiving, using one or more processors, a present record having values for a plurality of input variables;
• creating an input value for each of the plurality of input variables based on monotonic relationships assigned to the plurality of input variables;
• determining a plurality of predicted values using the following iterative process, where i starts at 1:
• (A) creating an i-th record by incorporating the values in the present record and substituting the value for the i-th input variable in the present record for the corresponding input value;
• (B) generating an i-th predicted value using a machine learning algorithm executed by one or more processors, the machine learning algorithm utilizing the values for the plurality of input variables in the i-th record and the monotonic relationships assigned to the plurality of input variables;
• and
• (C) increasing i by one and repeating steps (A)-(C) until i reaches the total number of the plurality of input variables;
• and
• generating at least one record level reason code based on the plurality of predicted values and a predicted value for the present record.

A system for automatically generating record level variable importance using a machine learning model, the system comprising:

• at least one processor;
• and
• one or more computer-readable media having computer-executable instructions embodied thereon, wherein, upon being executed by the at least one processor, the computer-executable instructions cause the at least one processor to:
• (A) assign a monotonic relationship to each of a plurality of input variables, each monotonic relationship providing a directional constraint;
• (B) receive a plurality of input values corresponding to the plurality of input variables;
• (C) determine a target variable using a machine learning algorithm executed by the at least one processor, the machine learning algorithm utilizing the plurality of input values and the monotonic relationships assigned to the plurality of input variables;
• (D) create a plurality of input values by altering at least some of the input values according to the monotonic relationships assigned to their corresponding input variables;
• (E) determine a plurality of target values using the machine learning algorithm executed by the one or more processors, the machine learning algorithm utilizing the plurality of input values, the plurality of input values, and the monotonic relationships assigned to the plurality of input variables, wherein each model is determined using one input value and each of the plurality of input values except for the input value that was altered to create that input value;
• and
• (F) determine record level variable importance for the interpretable machine learning model based on the plurality of target values and the target variable.

One or more non-transitory computer-readable storage media having computer-executable instructions embodied thereon, the computer-executable instructions configured to cause at least one processor executing the instruction to perform the following:

• assign a monotonic relationship to each of a plurality of input variables, each monotonic relationship providing a directional constraint;
• receive, a plurality of input values for a record, the input values corresponding to the plurality of input variables;
• and
• determine a target value using a machine learning algorithm executed by the at least one processor, the machine learning algorithm utilizing the plurality of input values and the monotonic relationships assigned to the plurality of input variables.