Method for generating rulesets using tree-based models for black-box machine learning explainability

One or more non-transitory machine-readable media storing instructions that, when executed by one or more processors, cause:

• a machine learning (ML) model generating a plurality of classifications, wherein each classification associates a distinct example of a plurality of examples with a label of a plurality of labels;
• defining a predefined mapping of a sequence of rule levels to a sequence of numeric fractions;
• after said defining the predefined mapping of the sequence of rule levels, obtaining a decision tree that, based on the plurality of classifications, contains a plurality of tree nodes that are logically arranged into a sequence of tree levels, wherein:
• each tree node of the plurality of tree nodes comprises:
• one or more labels of the plurality of labels, a condition that identifies a feature of a plurality of features of the plurality of examples, and a split value for the feature, and
• when the tree node has child tree nodes, then the feature and the split value that are identified by the condition of the tree node are set to maximize, based on the plurality of classifications, information gain of the child tree nodes;
• associating, based on said sequence of numeric fractions of the predefined mapping of the sequence of rule levels, each rule level of said sequence of rule levels with a respective portion of said sequence of tree levels;
• generating, based on said sequence of numeric fractions of the predefined mapping of the sequence of rule levels, a candidate plurality of decision rules by traversing the decision tree, wherein:
• each decision rule of the candidate plurality of decision rules contains a label of the plurality of labels and a condition of at least one tree node of the plurality of tree nodes,
• each decision rule of the candidate plurality of decision rules is assigned to a respective rule level of the sequence of rule levels;
• generating, by a rules optimizer, an optimized plurality of decision rules from the candidate plurality of decision rules.