Dynamic Ensemble Model for Imbalanced Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning models face challenges in anomaly detection due to imbalanced training data, where few anomalous events are relative to typical events, making it difficult to accurately train algorithms.
Innovation Solution
A dynamic ensemble machine learning model is trained using a subset of models selected based on their performance on similar training data, utilizing both distance and relative density metrics to ensure diversity and relevance, thereby improving accuracy by excluding poorly performing models and enhancing the influence of better-performing ones.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional machine learning models are trained on imbalanced anomaly detection data, then the training process can be completed, but the detection accuracy remains poor due to the scarcity of anomalous events
Solution Approach 1:
The patent combines multiple machine learning models into an ensemble system where each model processes the imbalanced training data differently. By merging diverse model predictions through weighted aggregation, the system achieves superior anomaly detection accuracy compared to individual models, effectively overcoming the limitations of scarce anomalous training examples
Solution Approach 2:
The patent dynamically adjusts model weights and selection criteria based on performance metrics calculated from available training data. By changing the parameters of model combination (weights, selection thresholds) rather than the models themselves, the system adapts to imbalanced data conditions and improves detection accuracy without requiring additional training data
2Adaptability or versatility
If a static ensemble of machine learning models is used, then model diversity is achieved, but the system cannot adapt to different data characteristics and performance varies across different data subsets
Solution Approach 1:
The patent transforms the static ensemble into a dynamic system where model selection and weighting change based on the characteristics of the input data. The system calculates similarity between test data and training data subsets, then selects models that performed best on similar historical data, enabling adaptation to different data characteristics without manual reconfiguration
Solution Approach 2:
The patent pre-calculates performance metrics for each model across different training data subsets during the training phase. This preliminary action creates a performance database that enables rapid model selection during inference, reducing the complexity of real-time adaptation while maintaining high adaptability to different data characteristics
3Measurement precision
If all trained machine learning models are used in the ensemble, then comprehensive coverage is achieved, but poorly performing models reduce overall accuracy and increase computational overhead
Solution Approach 1:
The patent extracts and removes poorly performing models from the ensemble by evaluating each model's contribution to overall accuracy. Models that consistently underperform or add noise to predictions are excluded from the final ensemble, reducing computational overhead while maintaining or improving detection accuracy. This selective extraction optimizes the balance between comprehensive coverage and efficiency
Data Source
AI summary
Described are techniques for using a dynamic ensemble model. The techniques including training a plurality of machine learning models on training data. The techniques further include identifying a similar subset of the training data that is similar to a dataset for evaluation. The techniques further include assembling a subset of models from the plurality of machine learning models based on performance of the subset of models on the similar subset of the training data. The techniques further include generating an output from the subset of models for the dataset for evaluation.