Augmenting time series data plays a crucial role in enhancing the generalization of classification models, especially in scenarios where labeled datasets are limited. By generating synthetic samples, data augmentation addresses the challenge of data scarcity, allowing models to learn more robust features and improve performance on unseen data. To support the development and evaluation of augmentation techniques, we introduce TATS: Toolbox for Time Series Data Augmentation, a comprehensive framework designed for generating synthetic time series. The toolbox includes ten augmentation methods and provides access to 17 benchmark time series datasets, enabling the design and comprehensive comparison of different augmentation approaches. TATS integrates classification pipelines based on Dynamic Time Warping (DTW) with k-Nearest Neighbors (k-NN), Long Short-Term Memory (LSTM) network, and four other classification methods, facilitating testing and benchmarking of new augmentation methods across various application domains. Implemented in MATLAB and available on GitHub ( https://github.com/maroszII/TATS ), TATS provides a standardized and extensible platform for advancing research in time series data augmentation.
Warchoł et al. (Mon,) studied this question.