Loan Pricing Model Weighting to Mitigate Demographic Disparities
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
AI/ML models generate disparate outputs based on training factors, leading to unfair impacts on demographically classified groups, necessitating a mechanism to mitigate these disparities under usage constraints.
Innovation Solution
A method involving the calculation of model weights and training a customized model using feature weight functions, sensitive labels, and historical data to generate loan prices that consider both profit and demographic fairness, combining multiple AI/ML models to optimize outputs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If multiple AI/ML models are used to generate loan prices, then profit optimization is improved, but output disparities and demographic unfairness worsen
Solution Approach 1:
The patent combines multiple AI/ML models into an ensemble system where each model's output is weighted and aggregated. This merging approach allows the system to leverage the profit-optimizing capabilities of multiple models while reducing individual model biases through diversification, thereby addressing both profit optimization and output consistency simultaneously
Solution Approach 2:
The system dynamically adjusts model weights based on performance metrics and uncertainty values. By changing the parameters (weights) assigned to each model's output, the system can optimize for profit while maintaining consistency by downweighting models that produce disparate or biased outputs
2Reliability
If model weights are calculated based on uncertainty values and quality parameters, then output reliability is improved, but system complexity worsens
Solution Approach 1:
The system performs self-evaluation by automatically calculating uncertainty values and quality parameters for each model based on their own performance metrics. This self-service approach improves reliability through continuous monitoring while avoiding the need for external complex evaluation systems
Solution Approach 2:
The system implements feedback loops where model outputs are evaluated against ground truth data, and the resulting quality metrics feed back into weight calculations. This feedback mechanism continuously improves reliability while maintaining manageable complexity through automated closed-loop control
3Object-affected harmful factors
If a customized model is trained using sensitive labels and historical data, then demographic fairness is improved, but training data requirements worsen
Solution Approach 1:
The system performs preliminary actions by pre-processing historical data to identify and incorporate sensitive labels before training the customized model. This preliminary preparation ensures that demographic fairness considerations are built into the training process from the outset, reducing bias without requiring excessive additional data
Solution Approach 2:
The training approach combines multiple data sources including historical loan data, sensitive demographic labels, and synthetic balanced data. This composite data strategy enables the model to learn fair patterns across demographic groups while managing data quantity requirements through strategic data combination
Data Source
AI summary
Various methods and processes, apparatuses or systems, and media for mitigating disparities between outputs of different AI/ML models that are subject to usage constraints are disclosed. The method includes: receiving uncertainty values and model quality-related parameter values that are associated with at least two models that are configured to generate a loan price for a loan applicant; receiving feature weight functions that relate to weights of target metrics; calculating model weights for each model; selecting a customized model based on the model weights; receiving a tabular set of personal data that includes individualized financial information and individualized demographic information associated with loan applicants; training the customized model by using the tabular set of personal data, the target metrics, a set of sensitive labels, and historical information that relates to outputs generated by the models; and using the trained model to generate a customized loan price for a loan applicant.