Rejection-Region Ensemble Learning for Credible, Efficient Predictions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing ensemble learning methods face challenges such as high computational cost, limitations in baseline model ranking and aggregation, especially for class-imbalanced data, and lack of credibility measurement in feature patterns, leading to inefficient predictive performance.
Innovation Solution
The method involves training baseline machine learning models on diverse data types, defining rejection regions based on performance metrics, generating global rejection regions, and updating models until optimal performance is achieved, using a meta-model for flexible aggregation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional static ensemble learning is used, then implementation is simpler, but predictive performance is lower
Solution Approach 1:
The patent implements dynamic ensemble selection where the system adaptively selects and weights baseline models based on their performance on specific test samples. The rejection region dynamically adjusts which models contribute to predictions, transforming the static ensemble into a dynamic, sample-specific selection mechanism that optimizes predictive performance while managing complexity.
Solution Approach 2:
The patent applies local quality by allowing different baseline models to have different weights or selection statuses depending on the specific test sample. The rejection region is sample-specific, meaning each test sample can have a different set of active baseline models, enabling localized optimization for different data patterns and feature combinations.
2Reliability
If selective ensemble learning with dynamic model selection is implemented, then predictive performance improves, but computational cost increases
Solution Approach 1:
The patent extracts and removes underperforming baseline models from the ensemble for specific test samples by defining rejection regions. Instead of running all baseline models on all samples, the system identifies and excludes models whose predictions fall within rejection regions, thereby reducing computational cost while maintaining or improving predictive performance on the remaining samples.
Solution Approach 2:
The patent applies partial action by selectively applying only the necessary subset of baseline models to each test sample rather than running the full ensemble. The rejection mechanism allows the system to do less computation (exclude certain models) for samples where it's not needed, while maintaining full ensemble capability for samples where it provides value.
3Reliability
If predictions are made for all data samples, then coverage is complete, but model credibility is not measured
Solution Approach 1:
The patent segments the prediction space into acceptance regions and rejection regions for each baseline model and for the ensemble. This segmentation allows the system to distinguish between samples where predictions are credible (in acceptance regions) and samples where predictions are unreliable (in rejection regions), thereby measuring model credibility while maintaining overall prediction coverage through selective application.
4Reliability
If baseline models are ranked and aggregated using traditional methods, then aggregation is straightforward, but performance is limited especially for class-imbalanced data
Solution Approach 1:
The patent changes the parameters used for model ranking and aggregation by incorporating rejection region information into the selection and weighting of baseline models. Instead of using traditional performance metrics alone, the system uses rejection region characteristics (which samples are rejected by which models) as additional parameters for determining model contribution, thereby improving performance on class-imbalanced data through more nuanced model aggregation.
Data Source
AI summary
Methods, systems, and computer program products are provided for ensemble learning. An example system includes at least one processor configured to: (i) generate a rejection region for each baseline model of a set of baseline models (ii) generate a global rejection region based on the rejection regions of each baseline model; (iii) train an ensemble machine learning model; (iv) update, based on a baseline model predictive performance metric for each baseline machine learning model, the set of baseline machine learning models; and (iv) repeat (i)-(iv) until there is a single baseline model in the set of baseline models or a predictive performance or global acceptance ratio of the ensemble model satisfies a threshold.