Cost-Sensitive Fraud Detection Model Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems fail to accurately detect fraudulent transactions during the early stages of a newly launched system, and they do not account for the varying costs of misclassification errors in financial transactions, leading to potential losses.
Innovation Solution
A system that uses a processor to identify overlapping features between first and second training data sets to generate third training data, and employs a cost-sensitive loss function to differentiate misclassification error costs, thereby improving the accuracy of fraudulent transaction detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If transfer learning is used to train a machine learning model for a new system using data from a densely populated system, then the model can be trained with limited historical data available at launch, but the model is prone to errors during the earlier stages of the new system
Solution Approach 1:
The patent applies preliminary action by pre-training a source machine learning model on a densely populated source system before using its learned features to train the target model. This preliminary training on abundant source data prepares the model to handle the target system's limited historical data effectively, reducing errors during early deployment stages
Solution Approach 2:
The patent uses an intermediary approach by introducing a source machine learning model as a mediator. The source model is trained on abundant historical data from a source system, and its learned features serve as an intermediary representation that helps the target model learn from limited target historical data, bridging the data gap between source and target systems
2Device complexity
If current machine learning models assume equal cost for all misclassifications, then the model training process is simplified, but the system fails to accurately determine the actual cost of misclassification errors in fraud detection
Solution Approach 1:
The patent applies local quality by introducing cost-sensitive loss functions that assign different misclassification costs to different classes (e.g., higher cost for false negatives in fraud detection). This allows the model to account for varying local conditions and consequences of errors in different regions of the decision space, improving measurement precision of misclassification costs
Solution Approach 2:
The patent changes the parameter of misclassification cost from a uniform value to variable values based on the specific class and context. By incorporating cost-sensitive loss functions with adjustable cost parameters, the model can accurately reflect the actual business costs of different types of errors without excessive complexity
Data Source
AI summary
Systems and methods for training a machine learning model are disclosed. A new machine learning model for a new system is trained using portions of source data used to train a well-established machine learning model that solves a different but related problem compared to the new machine learning model. Features required to train the new machine learning model may be compared to features in data samples of the source data to determine the portion of source data that can be used as training data to train the new machine learning model. The training data may then be used to train the new machine learning model without requiring a large set of training data that is unavailable for the new system.