Self-Calibrating Outlier Model for Fraud Detection

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Traditional fraud detection models rely heavily on historical data and struggle with adapting to changing fraud patterns and environments, especially when high-quality historical data is unavailable or fraud patterns evolve rapidly.

Innovation Solution

An adaptive outlier model combining a self-calibrating outlier model and an adaptive cascade model, which reduces dependency on historical data and enables learning of changing fraud patterns in real-time production environments by using transaction data and feedback loops for continuous updating.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional supervised fraud modeling is used with historical data, then model accuracy is improved, but the model cannot adapt to changing fraud patterns and environments

Engineering Contradiction:

Improvefraud detection accuracyVSAvoidadaptability to changing fraud patterns

Core Design Contradiction:

Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent implements dynamic adaptability by enabling the fraud detection model to continuously learn and update from production environment feedback. The system transitions from static historical data modeling to dynamic real-time learning, where the model adapts its parameters and patterns based on ongoing transaction analysis and fraud case outcomes, allowing it to respond to evolving fraud tactics while maintaining detection accuracy

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The patent establishes a feedback loop where fraud detection outcomes and case investigation results are fed back into the model for continuous improvement. The system uses labeled fraud cases from human analysts to retrains the model, creating a closed-loop system that continuously refines its detection capabilities based on real-world performance, thereby maintaining both accuracy and adaptability

Inventive Principle:

Principle #23Feedback

2Adaptability or versatility

If self-calibrating unsupervised models are used without historical data, then adaptability to new environments is improved, but fraud detection accuracy decreases

Engineering Contradiction:

Improveability to operate without historical dataVSAvoidfraud detection accuracy

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent applies preliminary action by pre-configuring the unsupervised model with calibration parameters and detection thresholds before deployment. The system performs initial calibration using available data to establish baseline fraud patterns, enabling it to operate effectively in new environments from the start while maintaining reasonable detection accuracy even before extensive real-time learning occurs

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent implements self-service through automated calibration and adaptation mechanisms that allow the model to self-adjust its parameters based on production environment feedback. The system automatically retrains on new fraud cases and adjusts its detection thresholds without requiring manual intervention, enabling it to maintain and improve accuracy autonomously while adapting to new environments

Inventive Principle:

Principle #25Self-service

3Adaptability or versatility

If model retraining is performed frequently to adapt to changing fraud patterns, then adaptability is improved, but deployment time and computational resources increase

Engineering Contradiction:

Improveresponse to changing fraud environmentVSAvoidmodel retraining and deployment time

Core Design Contradiction:

Adaptability or versatilityVSLoss of time

Solution Approach 1:

The patent implements periodic action by scheduling model retraining at optimized intervals based on fraud pattern change detection. Rather than continuous retraining, the system monitors for significant deviations in fraud patterns and triggers retraining only when necessary, balancing adaptability with computational efficiency and deployment time constraints

Inventive Principle:

Principle #19Periodic action

Solution Approach 2:

The patent applies parameter changes by adjusting model retraining frequency and calibration parameters based on environmental conditions. The system dynamically modifies its learning rate, retraining triggers, and calibration thresholds to optimize the balance between adapting to changing fraud patterns and minimizing deployment time and computational resource consumption

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS8041597B2Self-calibrating outlier model and adaptive cascade model for fraud detection

Publication Date: 2011.10.18 FAIR ISAAC & CO INC

AI summary

A system and method for detecting fraud is presented. A self-calibrating outlier model is hosted by a computing system. The self-calibrating outlier model receives transaction data representing transactions, and is configured to calculate transaction-based variables, profiles and calibration parameters, and to produce a score based on the transaction data according to the transaction-based variables, profiles and calibration parameters. An adaptive cascade model is also hosted by the computing system, and is configured to generate a secondary score for the transaction data based on profile information from the variables and/or profiles calculated by the self-calibrating outlier model, and based on a comparison with labeled transactions from a human analyst of historical transaction data.