Tree-Based Model Fairness Bounds Through Path-Similarity Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems lack robustness and customization for evaluating sensitivity and fairness bounds of tree-based models, particularly in determining upper and lower bounds on fairness during the production stage, especially when dataset records are similar to each other.
Innovation Solution
A system comprising a fairness component, identification component, and evaluation component that identifies root-to-leaf paths, removes records with similar paths, and evaluates sensitivity based on first and second fairness metrics, with data perturbation to monitor data drift and alert when fairness thresholds are breached.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If records with similar root-to-leaf paths are removed from the dataset, then the accuracy of fairness evaluation is improved, but the quantity of data is reduced
Solution Approach 1:
The patent extracts and removes records with similar root-to-leaf paths from the dataset. The identification component detects records sharing identical paths through the tree-based model, and the removal component eliminates these redundant records. This extraction process isolates the problematic similar records while preserving the rest of the dataset, thereby improving fairness evaluation accuracy without completely discarding valuable data.
Solution Approach 2:
The patent segments the dataset based on root-to-leaf path characteristics. By grouping records according to their traversal paths through the tree-based model and then removing only the redundant segments (similar paths), the system maintains diversity in the remaining data while eliminating evaluation inaccuracies caused by path similarity.
2Measurement precision
If data perturbation is applied to monitor data drift, then the sensitivity and fairness bounds are accurately determined, but the complexity of the evaluation process increases
Solution Approach 1:
The patent applies data perturbation as a preliminary action before fairness evaluation. By intentionally modifying data attributes and observing model responses in advance, the system determines sensitivity and fairness bounds upfront. This preliminary characterization simplifies the main evaluation process by establishing reference thresholds that guide subsequent fairness assessments.
Solution Approach 2:
The patent implements feedback mechanisms where data perturbation results feed into the fairness evaluation process. The system monitors how perturbations affect model predictions and uses this feedback to adjust fairness assessments and identify data drift. This closed-loop approach systematically determines sensitivity bounds while managing complexity through structured feedback analysis.
3Reliability
If the system monitors data drift and alerts when fairness thresholds are breached, then the reliability of fairness monitoring is improved, but the time required for continuous evaluation increases
Solution Approach 1:
The patent replaces continuous manual fairness monitoring with an automated alerting system. The system establishes fairness thresholds and automatically detects when data drift causes threshold breaches, substituting continuous human evaluation with automated computational monitoring. This maintains high reliability through consistent threshold checking while reducing time loss by eliminating manual intervention requirements.
Data Source
AI summary
One or more systems, devices, computer program products and/or computer-implemented methods of use provided herein relate to a process to facilitate abnormal document self-discovery. A system can comprise a memory that stores computer executable components, and a processor that executes the computer executable components stored in the memory, wherein the computer executable components can comprise a fairness component, an identification component, a removal component, and an evaluation component. The fairness component can receive a dataset for a tree-based model and calculates a first fairness, and the identification component can identify root to leaf paths in the tree-based model for one or more records of the dataset and one or more corresponding perturbed records. The removal component can remove at least one record of the one or more records having similar root to leaf paths to the other records of the one or more records.