Rule-Based Fraud Detection via Decision Tree Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current fraud detection systems, particularly in e-transfer applications, face challenges in achieving high precision due to heavily imbalanced fraud data, where less than 1% of all traffic is fraudulent, leading to inefficiencies in identifying fraudulent activities.
Innovation Solution
A rule-based machine learning approach that involves sampling user data, combining it with confirmed fraudulent data, and using a decision tree model to transform characteristics into numerical aggregates, followed by feature engineering and iterative decision tree processes to identify highly correlated fraud characteristics, generating a report that meets an accuracy precision threshold.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional fraud detection methods are used on heavily imbalanced data (less than 1% fraud), then the system processes all traffic, but the precision of fraud detection remains low due to the imbalance
Solution Approach 1:
The patent segments the fraud detection process into multiple iterative decision tree models. Each model focuses on a specific subset of fraudulent characteristics identified in previous iterations, allowing the system to progressively refine detection precision on different segments of the imbalanced data without being overwhelmed by the overall volume of non-fraudulent traffic
Solution Approach 2:
The patent performs preliminary actions by first identifying and isolating confirmed fraudulent data points before running the full dataset through decision tree models. This preliminary segmentation of fraudulent cases allows subsequent models to focus computational resources on learning from actual fraud patterns rather than being diluted by the majority non-fraudulent traffic
2Measurement precision
If the decision tree model is run on the full dataset to identify fraud characteristics, then comprehensive fraud patterns can be detected, but the computational complexity and processing time increase
Solution Approach 1:
The patent extracts and isolates specific fraudulent characteristics from each decision tree iteration, taking out only the high-value insights (nodes with high class 1 Gini index) for further analysis. This extraction approach reduces complexity by focusing subsequent iterations on specific fraud indicators rather than re-analyzing the entire feature space
Solution Approach 2:
The patent applies partial action by running decision tree models on progressively refined subsets of data in each iteration rather than repeatedly processing the full dataset. Each iteration focuses on a specific portion of fraud characteristics, reducing overall computational complexity while maintaining comprehensive fraud detection through the cumulative effect of multiple partial analyses
3Measurement precision
If multiple iterations of the decision tree model are performed to achieve high precision, then fraud detection accuracy improves, but the processing time and computational resources increase
Solution Approach 1:
The patent uses partial action by performing multiple iterations of decision tree models on progressively refined data subsets rather than running exhaustive analyses on the full dataset each time. Each iteration processes only the relevant portion of data needed to refine specific fraud characteristics, reducing total processing time while achieving high precision through cumulative refinement
Solution Approach 2:
The patent performs preliminary actions by pre-identifying and isolating confirmed fraudulent cases before the iterative modeling process. This preliminary preparation allows subsequent iterations to focus computational resources on learning from actual fraud patterns rather than wasting time analyzing non-fraudulent traffic, thereby reducing overall processing time while maintaining high accuracy
Data Source
AI summary
Methods and systems for rule-based machine learning for precise fraud detection. One system includes an electronic processor configured to determine, via a decision tree, a first subset of datasets of an aggregate dataset collection generated using a rule-based model. The electronic processor is also configured to select a third collection of datasets, each dataset included in the third collection of datasets associated with a user characteristic associated with fraud. The electronic processor is also configured to determine, via the decision tree, a second subset of datasets of the third collection of datasets, each dataset included in the second subset of datasets associated with a second set of user characteristics associated with fraud. The electronic processor is also configured to, in response to determining that an accuracy score associated with the second set of user characteristics satisfies a threshold, generate and transmit a report for display.