Predictive Model Bias Removal Through Node Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Predictive performance models often generate biased outcomes despite excluding protected data, as biases can emerge during machine processing, particularly in neural networks or decision trees, leading to unethical or illegal results.
Innovation Solution
A system and process that utilize a hardware processor to receive input data, generate an artificial intelligence neural network, determine validity and bias thresholds, score nodes based on predictive bias, and generate a new neural network excluding biased parameters to mitigate bias in predictive models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If protected data is excluded from input to predictive models, then legal and ethical compliance is improved, but bias in model outputs still occurs due to unprotected data combinations
Solution Approach 1:
The system performs preliminary bias detection and variable identification before final model deployment. It analyzes variable combinations, interaction terms, and hidden layers to pre-identify bias sources, then applies corrections in advance to prevent biased outcomes while maintaining compliance with protected data exclusions
Solution Approach 2:
The system introduces intermediary variables and proxy measurements that mediate between unprotected input variables and model outputs. These intermediaries capture the relationship between variables without directly using protected data, allowing the model to maintain predictive accuracy while avoiding direct bias from protected attributes
2Measurement precision
If initial inputs are individually unbiased, then data quality is improved, but bias emerges during machine processing in neural networks or decision trees
Solution Approach 1:
The system implements feedback loops that continuously monitor model outputs and intermediate processing stages. It detects bias emergence during neural network processing or decision tree operations, then feeds this information back to adjust variable weights, remove problematic combinations, or modify processing parameters to eliminate the bias
Solution Approach 2:
The system segments the model processing into analyzable components, examining each layer of the neural network or each node in the decision tree separately. This segmentation allows identification of specific processing stages where bias emerges, enabling targeted interventions at the source rather than treating the entire model as a black box
3Productivity
If variable combinations are analyzed recursively to improve prediction accuracy, then model performance is improved, but biased combinations are introduced
Solution Approach 1:
The system performs partial analysis of variable combinations by examining subsets of variables and their interactions up to a certain depth or complexity threshold. It identifies and includes only those combinations that meet bias criteria, excluding problematic higher-order interactions while retaining beneficial lower-order relationships, thus achieving acceptable accuracy without excessive bias introduction
Data Source
AI summary
A hardware processor can receive a set of input data individually describing a particular asset associated with an entity. The hardware processor can receive sets of inputs individually responsive to a respective subset of queries. The hardware processor can generate a predictive model using the set of input data. The hardware processor can calculate predictive outcomes individually associated with a respective user by applying the predictive model to each respective set of inputs of the sets of inputs. The hardware processor can generate a list ranked according to the predictive outcomes for the particular asset.