Prediction Target Clustering for More Accurate Credit Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated prediction mechanisms struggle with accuracy issues, particularly when modifying them to address underperforming categories, as they fail to distinguish between individual entities within a category that performed well and those that did not, leading to suboptimal adjustments.
Innovation Solution
The technique involves identifying clusters of individual prediction targets (IPTs) into UE cores and boundary PEs, determining the DIFF-SET variables and critical ranges, and using these to fine-tune the automated prediction mechanism by adjusting the decision tree based on these clusters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If automated prediction mechanism modifies the decision tree to address underperforming categories, then the overall accuracy improves, but the distinction between individual performing and underperforming entities within categories is lost
Solution Approach 1:
The patent segments the population of individual prediction targets into distinct clusters based on their feature values and performance characteristics. By dividing the data into multiple clusters with different performance patterns, the system preserves individual entity distinctions while still addressing category-level underperformance through targeted cluster-specific adjustments to the decision tree.
Solution Approach 2:
The patent applies local quality by creating cluster-specific performance metrics and adjustment strategies rather than uniform category-level treatments. Each cluster receives customized decision tree adjustments based on its specific performance characteristics, allowing the system to maintain sensitivity to individual entity performance while improving overall prediction accuracy for underperforming segments.
2Measurement precision
If automated prediction mechanism uses manual adjustments to create new versions, then prediction accuracy can be improved, but the modification process becomes complicated and error prone
Solution Approach 1:
The patent implements self-service by enabling the automated prediction mechanism to automatically identify underperforming clusters, generate cluster-specific adjustments, and update the decision tree without requiring manual intervention. The system autonomously monitors performance metrics, segments data into clusters, and applies targeted modifications, thereby improving accuracy while eliminating the complexity and error-proneness of manual versioning processes.
Solution Approach 2:
The patent incorporates feedback mechanisms that continuously monitor the performance of individual prediction targets and cluster-level metrics. This feedback loop automatically triggers cluster identification and decision tree adjustments when underperformance is detected, creating a self-correcting system that improves accuracy through automated iterative refinement rather than complicated manual versioning.
3Ease of operation
If automated prediction mechanism treats all entities in a category uniformly, then the decision process is simplified, but the accuracy suffers due to inability to distinguish performing from underperforming entities
Solution Approach 1:
The patent segments entities within categories into distinct clusters based on performance metrics and feature patterns. This segmentation maintains operational simplicity by automating the segmentation process and providing clear cluster definitions, while simultaneously improving accuracy by enabling differentiated treatment of performing versus underperforming entities through cluster-specific decision tree adjustments.
Solution Approach 2:
The patent introduces dynamics by making the segmentation and treatment approach adaptive rather than static. The system dynamically identifies clusters based on current performance data and automatically adjusts decision tree parameters for each cluster, maintaining simplicity through automation while achieving the accuracy benefits of differentiated entity treatment.
Data Source
AI summary
Techniques are provided for forming clusters of individual prediction targets (IPTs). An initial prediction target is a target for which an automated prediction has been generated. IPTs may be, for example, borrowers to which a lending entity has extended loans based on predictions generated by a credit policy. Each cluster includes (a) a “core” of underperforming entities (UEs), and (b) a set of boundary performant entities (PEs). The UEs that belong to the UE core of a cluster are “similarly situated” relative to the values of their features. For example, in the context where the IPTs are borrowers, the UEs at the core of a cluster may correspond to defaulting borrowers that had similar bureau data, lending entity data, and borrower data. The boundary performant entities of the cluster may be borrowers that have not defaulted, but had similar credit qualifications as the UEs of the cluster. Having formed these clusters, the clusters may be used in a variety of ways, including but not limited to improving the accuracy of the credit model, identifying potentially problematic future borrowers, generating visualizations that illustrate the relative importance of clusters of defaulting borrowers, etc.