Data Science Model Bias Mitigation via Input Variable Transformations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data science models used for business decision-making often incorporate bias that unfairly favor one subpopulation over another, making it difficult to comply with regulations and requiring inefficient trial-and-error methods to mitigate bias while preserving performance.
Innovation Solution
A computing platform evaluates an initial model object for bias, identifies biased input variables, applies transformations with unknown parameters, and uses Bayesian optimization to produce a post-processed model object with reduced bias, enabling efficient and flexible bias mitigation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional trial-and-error methods are used to mitigate bias in data science models, then bias reduction can be achieved, but the process is inefficient and resource-intensive
Solution Approach 1:
The patent applies parameter changes by transforming input variables through mathematical transformations (e.g., logarithmic, exponential, polynomial) to alter their statistical properties and reduce bias. The system automatically adjusts transformation parameters based on bias evaluation metrics, enabling efficient bias mitigation without trial-and-error model retraining.
Solution Approach 2:
The patent replaces the mechanical trial-and-error process with an automated computational system that uses Bayesian optimization and bias evaluation algorithms. This substitution of manual iterative processes with automated computational methods significantly improves productivity while achieving reliable bias reduction.
2Reliability
If bias mitigation transformations are applied to input variables, then fairness improves, but model performance may deteriorate
Solution Approach 1:
The patent implements dynamics by creating a flexible framework where transformation parameters can be dynamically adjusted based on bias evaluation results. The system automatically tunes transformation strength and type to achieve fairness while preserving performance, allowing adaptive balance between these competing objectives.
Solution Approach 2:
The system changes parameters of input variables through controlled transformations, adjusting them based on bias metrics while maintaining their predictive capability. This parameter transformation approach enables simultaneous optimization of fairness and performance by finding optimal transformation parameters that balance both objectives.
3Reliability
If multiple input variables are transformed to reduce bias, then fairness improves, but model complexity increases
Solution Approach 1:
The patent applies segmentation by evaluating and transforming input variables independently based on their individual bias contributions. The system identifies which variables require transformation and applies appropriate transformations to those specific variables, rather than uniformly transforming all variables, thus managing complexity through targeted intervention.
Solution Approach 2:
The patent introduces transformation functions as intermediary components between the raw input variables and the model. These transformation functions act as mediators that adjust the input data to reduce bias while maintaining a relatively simple overall model structure, as the transformations are applied at the data preprocessing level rather than increasing model architecture complexity.
Data Source
AI summary
A computing platform may be configured to (i) train an initial model object for a data science model using a machine learning process, (ii) determine that the initial model object exhibits a threshold level of bias, and (iii) thereafter produce an updated version of the initial model object having mitigated bias by (a) identifying a subset of the initial model object's set of input variables that are to be replaced by transformations, (b) producing a post-processed model object by replacing each respective input variable in the identified subset with a respective transformation of the respective input variable that has one or more unknown parameters, (c) producing a parameterized family of the post-processed model object, and (d) selecting, from the parameterized family of the post-processed model object, one given version of the post-processed model object to use as the updated version of the initial model object for the data science model.