Ensemble Learning With Rejection Regions for Credible Predictions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing ensemble learning methods face challenges such as high computational cost and limitations in baseline model ranking and aggregation, especially for class-imbalanced data, and lack credibility in model predictions due to unmeasured feature patterns.
Innovation Solution
A method for ensemble learning with rejection, involving training baseline models on diverse data types, generating rejection regions, and processing samples outside these regions to improve predictive performance, followed by training an ensemble model on these samples, optimizing rejection boundaries using algorithms like Grid search or Bayesian optimization, and flagging samples within the rejection region as unreliable.
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 which baseline models to use for each prediction based on their performance on training data. This dynamic approach allows the system to achieve higher predictive performance by using only the most competent models for each specific prediction task, rather than statically combining all models equally.
Solution Approach 2:
The patent applies local quality by evaluating each baseline model's competence locally on specific training samples and making selection decisions based on sample-specific performance characteristics. This allows different models to be selected for different regions of the feature space, optimizing predictive performance for local data patterns.
2Reliability
If selective ensemble learning with model ranking is used, then predictive performance improves, but computational cost increases
Solution Approach 1:
The patent extracts and removes underperforming baseline models from the ensemble based on their competence evaluation. By identifying and eliminating models that do not meet performance thresholds, the system reduces computational cost by avoiding unnecessary predictions from weak models while maintaining high predictive performance through selective use of competent models.
Solution Approach 2:
The patent applies partial action by not requiring all baseline models to participate in every prediction. Instead, only a subset of models that demonstrate sufficient competence on the training data are selected for making predictions, reducing the overall computational burden while maintaining or improving predictive accuracy.
3Reliability
If predictions are made for all data samples, then coverage is complete, but model credibility is not measured
Solution Approach 1:
The patent uses training data samples as disposable evaluation objects to assess model competence. Each training sample is used to evaluate whether baseline models are credible for making predictions on similar test samples. This allows the system to measure model credibility without requiring additional expensive validation processes.
Solution Approach 2:
The patent implements feedback by using training data to evaluate baseline model performance and using this evaluation to determine which models should be selected for making predictions. The training data provides feedback on model competence, which then informs the selection process for test predictions, ensuring that only credible models are used.
4Adaptability or versatility
If baseline models are trained on diverse data types, then model robustness improves, but training complexity increases
Solution Approach 1:
The patent segments the training process by training each baseline model on the complete diverse training dataset independently. This segmentation allows each model to develop its own specialized capabilities while maintaining robustness through exposure to diverse data types. The segmentation of training simplifies the overall process compared to coordinated multi-model training approaches.
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.