The Impact of Data Augmentation on Time Series Classification Models: An In-Depth Study with Biomedical Data

Bikram De, Mykhailo Sakevych, Vangelis Metsis

Abstract: Data augmentation is the practice of applying various transformations to existing data to increase their size and diversity without collecting new data. While augmentation strategies are widely recognized and implemented in image-based deep learning (DL) workflows, the degree to which they are effective in the time series domain is unclear. This paper experimentally evaluates the utility of various common time series augmentation techniques, especially those relevant to the medical sector where data limitations are prevalent. We thoroughly examine popular time series augmentation and synthetic data generation methods to evaluate their effectiveness in downstream classification tasks, encompassing both traditional and DL-based approaches. This research aims to offer insights into the applicability and efficacy of data augmentation strategies in improving model generalization and mitigating data scarcity challenges, with a focus on biomedical time-series data.

External IDs:dblp:conf/aime/DeSM24