Time Series Data Augmentation for Imbalanced Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Time series data analysis using machine learning faces challenges due to imbalanced data distribution, where insufficient data for minority classes, such as fault data, leads to overfitting and reduced prediction accuracy.
Innovation Solution
The proposed solution involves an apparatus and method for augmenting time series data by generating additional training data with varying time periods and applying data augmentation schemes based on column characteristics, such as numeric and categorical columns, to improve data balance and model generalization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a large amount of training data is collected to improve machine learning accuracy, then the quantity of data increases, but the data distribution becomes imbalanced with insufficient minority class data
Solution Approach 1:
The patent applies data augmentation techniques to create synthetic copies of minority class samples. Specifically, it uses methods like SMOTE (Synthetic Minority Over-sampling Technique) to generate artificial fault data by interpolating between existing minority class samples, thereby increasing the amount of minority class training data without requiring additional real-world fault occurrences.
Solution Approach 2:
The patent transforms the training process by changing parameters such as class distribution balance. It applies oversampling to minority classes and undersampling to majority classes, fundamentally altering the data distribution parameters to achieve balanced training sets that improve prediction accuracy for imbalanced time series data.
2Quantity of substance
If data augmentation is applied to increase minority class data, then the balance of data distribution improves, but the complexity of the processing increases
Solution Approach 1:
The patent segments the data processing into distinct modules: one for detecting class imbalance, another for applying appropriate augmentation techniques, and a third for validation. This modular segmentation allows each component to handle specific aspects of the imbalance problem independently, reducing overall system complexity while maintaining effectiveness.
Solution Approach 2:
The system automatically detects data distribution imbalances and selects appropriate augmentation strategies without manual intervention. The algorithm self-adjusts by analyzing class distributions and applying oversampling or undersampling techniques autonomously, reducing the complexity burden on users while maintaining balanced data processing.
3Device complexity
If traditional training methods are used with imbalanced data, then the processing is simple, but the model suffers from overfitting and reduced generalization ability
Solution Approach 1:
The patent applies data balancing techniques before the main training process begins. By pre-processing the training data to achieve balanced class distributions through augmentation or resampling, the system prevents overfitting from occurring during training, thereby improving generalization ability while maintaining relatively simple processing workflows.
Data Source
Figure 1
Figure 2
Figure 3A
AI summary
An apparatus and method for analyzing time series data on the basis of machine learning are provided. According to the disclosed embodiments, it is possible to effectively augment time series data, which is a target to be learned, according to characteristics of the time series data, thereby solving a problem of overfitting a machine learning model due to limited training data and a problem of deterioration of prediction accuracy due to imbalance of distribution of time series data and improving reliability of time series data analysis. In addition, according to the disclosed embodiments, it is possible to effectively set an optimal parameter for augmenting time series data.