False discovery rate difference evaluation metric

Scope

Copy link to section

The false discovery rate difference metric evaluates generative AI assets and machine learning models.

• Types of AI assets:
  • Prompt templates
  • Machine learning models
• Generative AI tasks: Text classification
• Machine learning problem type: Binary classification

Scores and values

Copy link to section

The false discovery rate difference metric score indicates the pervasiveness of false positives among all positive transactions for monitored and reference groups.

• Range of values: 0.0-1.0
• Best possible score: 0.0
• Ratios:
  • Under 0: Less false positives in monitored group
  • At 0: Both groups have equal odds
  • Over 0: Higher rate of false positives in monitored groups

Evaluation process

Copy link to section

To calculate the false discovery rate difference, confusion matrices are generated for the monitored and reference groups to identify the amount of false and true positives for each group. The false and true positive values are used to calculate the false positive rate for each group. The false positive rate of the reference group is subtracted from the false positive rate of the monitored group to calculate the false discovery rate difference.

Do the math

Copy link to section

The following formula is used for calculating the false discovery rate (FDR):

The following formula is used for calculating the false discovery rate difference:

Parent topic: Evaluation metrics