Synthetic Transaction Data Generation for Bias-Resistant ML Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning models trained on real-world transaction data suffer from quality issues, collection bias, tag uncertainty, and outcome assignment biases, leading to unreliable performance due to non-representative data points and defensive tagging, which can be exploited by bad actors.
Innovation Solution
A system generates synthetic data by determining archetype probability distributions, clustering transaction data points, removing non-representative points, and generating updated distributions to train machine learning models, thereby reducing bias and improving model robustness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If real world transaction data is used to train machine learning models, then the models can learn from actual transaction patterns, but the models produce unreliable results due to data quality issues, collection bias, and outcome assignment biases
Solution Approach 1:
The patent creates synthetic copies of transaction data that replicate the statistical properties and patterns of real data while eliminating quality issues. The system generates artificial transaction records that preserve multi-variate statistical distributions and transactional behavior patterns without inheriting the biases and errors present in real-world data collection
Solution Approach 2:
The patent extracts and removes non-representative data points from the training dataset by identifying and eliminating outliers that do not conform to established transactional behavior patterns. This extraction process separates the useful statistical patterns from the harmful biases and errors in the original data
2Reliability
If synthetic data is generated to eliminate bias and improve reliability, then model performance improves, but the data processing complexity increases
Solution Approach 1:
The patent performs preliminary clustering and pattern identification on the training data before model training to establish baseline transactional behavior patterns. By pre-processing the data to identify archetypes and statistical distributions, the system simplifies the subsequent synthetic data generation process while ensuring the synthesized data maintains realistic patterns
Solution Approach 2:
The patent transforms the training data by changing its parameters through synthetic generation, preserving statistical properties while modifying the underlying data source. The system adjusts parameters such as transaction amounts, frequencies, and patterns to match the statistical distributions of real data without copying the actual biased records
3Measurement precision
If all transaction data points are used for training, then the model learns from comprehensive data, but non-representative data points introduce noise and reduce model accuracy
Solution Approach 1:
The patent extracts and removes non-representative data points by comparing each data point against clustered transactional behavior patterns. Data points that fall outside acceptable deviations from their cluster archetypes are identified as outliers and eliminated, reducing noise while preserving the essential training signal
Solution Approach 2:
The patent applies different quality standards to different regions of the data space by clustering transactions into distinct behavioral patterns. Each cluster has its own archetypal characteristics, and data points are evaluated against their local cluster standards rather than a single global criterion, allowing nuanced filtering that preserves diversity while removing noise
Data Source
AI summary
A method may include generating synthetic data based on input data and training a machine learning model based on the synthetic data. The synthetic data may be generated by determining a plurality of data points representing an archetype probability distribution of a plurality of archetypes, clustering the plurality of data points into one or more clusters associated with transactional behavior patterns, generating a threshold metric representing a peak distribution density of the plurality of data points associated with a corresponding cluster, removing, from the plurality of data points, one or more non-representative data points to define a reduced set of the plurality of data points, generating an updated archetype probability distribution based at least on the reduced set of the plurality of data points, and generating representative transaction data based on the updated archetype probability distribution and threshold metric. Related methods and articles of manufacture are al so disclosed.