Time-Series Resampling for Rare-Event Class Imbalance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine-learning models trained on imbalanced datasets, particularly those with rare events, suffer from reduced accuracy due to skewed class distributions, leading to biased outcomes.
Innovation Solution
Generate synthetic time series for the minority class and/or reduce the number of time series for the majority class using neighboring time series based on a distance metric to balance the training dataset.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the training dataset uses real-world data with rare events, then the dataset reflects actual operational conditions, but the class distribution becomes heavily skewed towards the majority class
Solution Approach 1:
The patent creates synthetic copies of minority class time series by generating artificial data that mimics the characteristics of rare events. This copying approach allows the model to learn from sufficient examples of rare events without requiring actual increases in their occurrence in the training data, thereby resolving the class imbalance while maintaining reliability for rare event detection
Solution Approach 2:
The patent modifies parameters of existing time series data through mathematical transformations and noise additions to generate diverse synthetic samples. By changing parameters such as amplitude, frequency, and temporal characteristics while preserving the underlying patterns of rare events, the method creates a more balanced dataset that improves model accuracy without losing the essential features of minority class events
2Quantity of substance
If synthetic data is generated to balance class distribution, then class imbalance is reduced, but the complexity of data preprocessing increases
Solution Approach 1:
The patent divides the synthetic data generation process into distinct segments: identifying minority class samples, generating synthetic variants through parameter modifications, and integrating them with majority class data. This segmentation of the preprocessing workflow makes the complex task more manageable and systematic, reducing the overall complexity while achieving class balance
3Measurement precision
If more minority class samples are included in training, then model accuracy for rare events improves, but the risk of overfitting increases
Solution Approach 1:
The patent introduces dynamic variations in synthetic data generation by applying different noise levels, transformations, and parameter modifications to create diverse samples. This dynamic approach ensures that the model learns robust features that generalize well to unseen rare events, preventing overfitting while maintaining high classification accuracy through increased sample diversity
Data Source
AI summary
Class imbalance in a training dataset may negatively impact the accuracy of a machine-learning model in classifying rare events that are underrepresented in the training dataset. Training datasets comprising time-series data present a unique challenge. Accordingly, resampling techniques for up-sampling and/or down-sampling a training dataset of time series are disclosed. The up-sampling may respect the temporal correlation of time samples in the time series, while generating synthetic time series that mimic the feature values of time series belonging to the minority class. Down-sampling may be used to fine-tune the ratio of time series belonging to the minority class to the time series belonging to the majority class.