Shapelet-Preserving Bootstrapping For Time Series Data Augmentation

Omar Bahri, Peiyu Li, Pouya Hosseinzadeh, Soukaïna Filali Boubrahimi, Shah Muhammad Hamdi

Abstract: Data augmentation has proved to be extremely beneficial to the advancement of current state-of-the-art machine learning classification models. Not only does it allow balancing datasets collected in domains with rare positive events, but it also helps increase the size of small datasets to feed data-hungry deep learning models. More often than not, this results in a significant performance increase. However, data augmentation in the time series domain is still lagging compared to other domains such as image and text data. We propose BootShape, a novel data augmentation method for time series classification that bootstraps the residual components of original dataset instances to generate new time series patterns. By preserving the most important shapelets, BootShape is able to generate realistic data instances that respect the original class distributions. Using real-life datasets from the University of California Riverside (UCR) archive, we show that BootShape leads to higher performance gains compared to 6 state-of-the-art pattern mixing time series data augmentation methods.