Risk Determination ML Model Using Per-Horizon Claim Sets
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing predictive data analysis systems face inefficiencies and inaccuracies due to limited data and computational resources, particularly in generating risk measures for entities with insufficient direct data, such as caregivers, where indirect data from care recipients is abundant but not utilized effectively.
Innovation Solution
A risk determination machine learning model is trained using per-horizon historical claim sets from ground-truth entities with sufficient data, allowing it to generate predicted risk measures for entities outside the ground-truth subset, including caregivers, by leveraging hidden features and associated care recipient data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional predictive data analysis systems use direct data from predictive entities, then measurement precision is improved, but productivity deteriorates due to insufficient data availability for many entities
Solution Approach 1:
The patent introduces an intermediary approach by using care recipient data as a proxy mediator to infer caregiver risk profiles. Instead of directly analyzing caregiver claims (which are sparse), the system uses the data from care recipients to train machine learning models that can predict caregiver risks, thereby enabling analysis of entities with insufficient direct data
Solution Approach 2:
The system creates a copy of the data analysis process by training machine learning models on care recipient data patterns, then applying these models to predict caregiver risk measures. This copying approach allows the system to transfer knowledge from well-dataed entities (care recipients) to entities with limited data (caregivers)
2Productivity
If the system processes all candidate predictive entities, then productivity is improved, but measurement precision deteriorates due to insufficient data for entities outside the ground-truth subset
Solution Approach 1:
The patent segments the population of predictive entities into two groups: ground-truth entities with sufficient data (care recipients) and target entities with limited data (caregivers). The system trains models on the ground-truth segment and then applies them to the target segment, allowing comprehensive coverage while maintaining accuracy through the segmented training approach
Solution Approach 2:
The system changes the data parameters by transforming sparse claim count data into normalized risk ratios and then into machine learning model inputs. This parameter transformation allows the system to work with entities that have fewer than 100 claims by converting their sparse data into meaningful risk predictions through model-based parameter estimation
3Productivity
If the system uses machine learning models trained on limited ground-truth data, then productivity is improved for entities with insufficient data, but reliability deteriorates due to potential model inaccuracies
Solution Approach 1:
The system performs preliminary action by training machine learning models on ground-truth care recipient data before applying them to caregivers. This preliminary training phase establishes the model's predictive capabilities using reliable data, and the models are specifically designed to handle sparse data scenarios, thereby improving reliability when applied to entities with limited data
Solution Approach 2:
The patent incorporates feedback mechanisms by using ground-truth risk measures from care recipients to validate and refine the machine learning models. The models are trained on actual outcome data, allowing them to learn from feedback loops where predicted risks are compared with actual claims, thereby improving prediction accuracy and reliability
Data Source
AI summary
There is a need for more accurate and more efficient predictive data analysis steps/operations. This need can be addressed by, for example, techniques for efficient predictive data analysis steps/operations. In one example, a method includes generating, by a processor, utilizing a risk determination machine learning model and based at least in part on one or more hidden features of the first predictive entity, the predicted risk measure, and performing one or more prediction-based actions based at least in part on the predicted risk measure.