Efficient Time Series Augmentation Methods
Bo Liu, Zhenguo Zhang, Rongyi Cui
Abstract
With the emergence of deep learning and the improvement of computing ability, artificial intelligence has also made great progress. In the case of sufficient data, deep learning models can achieve very good results. However, in some tasks with scarce resources, the results of these models are somewhat unsatisfactory. In this regard, researchers have proposed some data augmentation technologies in the domain of computer vision and natural language processing. However, there is still a lack of effective methods for data augmentation applied to time series data. In this paper, we propose several methods to augment time series data, which are called AddNoise, Permutation, Scaling, Warping. Then, we verify these augmentation methods by two deep learning models, Fully Convolutional Neural Network (FCN) and ResNet, on real time series datasets. The experimental results show that these methods are useful to train the above two models in time series classification task.