Stratified Prediction Models for Heterogeneous Data Segments

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

As data sets become larger and more heterogeneous, traditional prediction algorithms that treat the entire data set as a single population experience decreased accuracy, as variables that are strong predictors for one segment of the population may have little effect on another segment, leading to suboptimal predictions.

Innovation Solution

The data is stratified using techniques adapted from principal component analysis, where prediction models are trained independently for each stratum, and predictions for a testing sample are made by combining per-stratum predictions weighted according to the probabilities of belonging to each stratum.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional prediction algorithms treat the entire data set as a single population, then the model complexity is low and ease of operation is high, but prediction accuracy deteriorates due to data heterogeneity

Engineering Contradiction:
Improveprediction accuracyVSAvoidmodel complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent divides the heterogeneous data set into multiple homogeneous strata or segments based on shared characteristics. Each stratum is then modeled separately with its own prediction model, allowing the system to capture population-specific patterns while maintaining overall accuracy. This segmentation resolves the contradiction by trading some model complexity for significant gains in prediction accuracy.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies local quality by creating prediction models with parameters tailored to each specific stratum rather than using a single global model. Each local model is optimized for its specific population segment, allowing the system to achieve high prediction accuracy for heterogeneous data by adapting model characteristics to local data properties.

Inventive Principle:
Principle #3Local quality

2Measurement precision

If prediction models are trained independently on each stratum, then prediction accuracy improves by accounting for heterogeneity, but the number of models and computational complexity increases

Engineering Contradiction:
Improveprediction accuracyVSAvoidcomputational efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent segments the training process into independent stratum-specific models, allowing parallel training and evaluation. This segmentation improves prediction accuracy by capturing stratum-specific relationships while enabling efficient computation through independent processing of each segment, resolving the trade-off between accuracy and computational efficiency.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent combines results from multiple stratum-specific prediction models into an integrated prediction framework. By merging the outputs of individual stratum models according to the testing sample's stratum probabilities, the system achieves high prediction accuracy while maintaining computational efficiency through result aggregation rather than complex joint modeling.

Inventive Principle:
Principle #5Merging (Combining)

3Adaptability or versatility

If the data set size increases to capture more population diversity, then the representativeness improves, but the heterogeneity increases and decreases prediction accuracy

Engineering Contradiction:
Improvepopulation representativenessVSAvoidprediction accuracy
Core Design Contradiction:
Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent segments the diverse population into homogeneous strata, allowing the data set to maintain high population representativeness through diversity while preserving prediction accuracy within each stratum. This segmentation enables the system to handle large, heterogeneous data sets effectively by ensuring that no single stratum dominates the overall prediction.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies local quality by creating stratum-specific prediction models that are optimized for each homogeneous group. This approach allows the system to maintain high prediction accuracy for each local population segment while the collection of strata collectively represents the entire diverse population, resolving the contradiction between representativeness and accuracy.

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP3809342A1Prediction models incorporating stratification of data
Publication Date: 2021.04.21 THE CHINESE UNIVERSITY OF HONG KONG
  • EP3809342A1 patent drawingFigure 1
  • EP3809342A1 patent drawingFigure 2
  • EP3809342A1 patent drawingFigure 3

AI summary

A Prism Vote process can be used to make predictions based on a data set that may be heterogeneous. The data set can be stratified, e.g., using techniques adapted from principal component analysis. A prediction model can be trained independently on each stratum of the data set. To make a prediction for a "testing" data sample, the prediction model for each stratum can be used to provide a per-stratum prediction of an outcome, and a per-stratum probability that the testing data sample belongs to each stratum can be determined. The predicted outcome (e.g., a probability of a particular outcome) can be computed from the per-stratum predictions weighted according to the per-stratum probabilities.