.png){kind=icon}
Return
Temporal 2D-cycle-generation framework for time series classification
.png){kind=icon}
PRE
AI
DOI：10.1016/j.asoc.2025.112778.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In time series classification tasks, most datasets have small data volumes or are inconvenient to collect. Therefore, we proposed a data augmentation framework based on two-dimensional time series data (CycleTime). CycleTime consists of an optimal dimension transformation method, an optimal data augmentation method, and an optimal classifier respectively. Through comparative experiments, Gramian angular difference field, optimized CycleGAN and FCN are selected to form the optimal framework. The experimental results indicate that CycleTime can effectively improve the classification accuracy of both EEG and HAR time series data. On average across five datasets, CycleTime improved the results by 1.4 % compared with the existing benchmark methods. CycleTime can provide reference value and theoretical basis for other tasks of time series.
Keywords:
Time series classification
Dimension transformation
Data augmentation
Generative adversarial network
Journal
IF:
6.6
Papers:
1.4W
Citations:
4.8W
Organization
No organization information available
Cited Papers
A deep learning framework for time series classification using Relative Position Matrix and Convolutional Neural Network.png)
NEUROCOMPUTING
NEUROCOMPUTINGIF6.5
PREAI
ShareSaveShareSave
Time series shapelets: a novel technique that allows accurate, interpretable and fast classification
OAAI
OAAIShareSaveShareSaveShareSaveShareSave
View more