Poisson Shelf Segmentation for Fraud Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for fraud prevention in transaction risk assessment lack accuracy and automation, particularly in identifying unusual patterns in call data from phone numbers, which are crucial for determining risk and managing future calls from potentially risky numbers.
Innovation Solution
The method employs a Poisson distribution to model count data, calculates a threshold value based on the mean and standard deviation, partitions the data, identifies 'shelves' representing abrupt variations, and defines outliers for risk assessment, applicable to both univariate and multivariate datasets.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional fraud detection methods are used, then the system can identify some risky transactions, but the accuracy is insufficient and false positives occur frequently
Solution Approach 1:
The patent segments call data into distinct groups or 'shelves' based on call frequency patterns. Each shelf represents a different behavioral category (e.g., normal callers, moderate callers, high-frequency callers). This segmentation allows the system to apply different risk assessment criteria to different segments, improving detection accuracy while reducing false positives by not treating all calls uniformly.
Solution Approach 2:
The patent transforms raw call count data into standardized parameters by fitting Poisson distributions to each shelf and calculating standardized residuals. This parameter transformation converts absolute call frequencies into relative deviations from expected patterns, enabling more accurate cross-comparison and risk assessment across different time periods and call volumes.
2Extent of automation
If manual risk assessment methods are used, then detailed analysis can be performed, but the process is not automated and requires significant human effort
Solution Approach 1:
The system performs self-service by automatically fitting Poisson distributions to data segments, calculating expected call frequencies, computing standardized residuals, and identifying outliers without human intervention. The algorithm autonomously determines which calls are risky based on statistical deviations from normal patterns, maintaining high precision while achieving full automation.
Solution Approach 2:
The patent incorporates feedback mechanisms where the system continuously monitors call patterns, updates Poisson distribution parameters based on observed data, and adjusts risk thresholds dynamically. This feedback loop allows the automated system to learn from historical data and improve its detection accuracy over time, matching or exceeding manual assessment quality.
3Ease of manufacture
If simple threshold-based methods are used, then the system is easy to implement, but it cannot accurately identify unusual patterns in call data
Solution Approach 1:
The patent applies preliminary action by first segmenting data into shelves and fitting Poisson distributions before performing outlier detection. This preparatory step establishes expected call frequency patterns for each segment, creating a foundation for more accurate subsequent analysis. The system pre-calculates standardized residuals and identifies potential outliers before final risk classification, improving precision while maintaining implementation simplicity.
Data Source
AI summary
Detecting fraudulent activity can be a complex, manual process. In this paper, we adapt statistical properties of count data in a novel algorithm to uncover records exhibiting high risk for fraud. Our method identifies shelves, partitioning data under the counts using a Student's t-distribution. We apply this methodology on a univariate dataset including cumulative results from phone calls to a customer service center. Additionally, we extend this technique to multivariate data, illustrating that the same method is applicable to both univariate and multivariate data.