Protected Class Model Evaluation via Dimensionality Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning systems lack transparency and tools to determine whether their outcomes disparately impact protected classes, making it difficult for businesses to ensure compliance with laws and regulations.
Innovation Solution
A model evaluation system that assesses machine learning models for disparate impact by using protected class information and model evaluation techniques, identifying variables that influence predictions and scores, and adjusting models to minimize adverse effects on protected groups.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If complex machine learning algorithms (such as neural networks, ensembled models) are used to improve predictive power, then model accuracy and predictive capability are improved, but model transparency and interpretability deteriorate
Solution Approach 1:
The patent introduces dimensionality reduction techniques as an intermediary layer between complex machine learning models and stakeholders. By transforming high-dimensional model features into lower-dimensional representations, the system maintains the predictive power of complex models while making them interpretable and evaluable for disparate impact analysis
Solution Approach 2:
The patent replaces direct inspection of complex model internals with dimensionality reduction-based evaluation methods. Instead of attempting to interpret complex neural networks or ensembled models directly, the system uses dimensionality reduction to create simplified representations that capture the essential discriminatory patterns
2Device complexity
If dimensionality reduction is applied to simplify model evaluation, then model transparency and interpretability are improved, but information loss may occur
Solution Approach 1:
The patent applies dimensionality reduction techniques that transform the parameter space of model features while preserving the essential discriminatory information. By carefully selecting reduction methods and parameters, the system maintains the ability to detect disparate impact patterns while reducing complexity
Data Source
AI summary
Systems and methods for model evaluation. A protected class model that satisfies an accuracy threshold is built by using: data sets for use by a modeling system being evaluated, and protected class membership information for each data set. A target for the protected class model is a protected class membership variable indicating membership in a protected class. Each predictor of the protected class model is a predictor of an evaluated model used by the modeling system. A target of the evaluated model is different from the target of the protected class model. Each predictor is a set of one or more variables of the data sets. For each predictor of the protected class model, a protected class model impact ranking value and a modeling system impact ranking value are determined.