Data augmentation using synthetic data for time series classification with deep residual networks

Data augmentation in deep neural networks is the process of generating data in order to reduce the variance of the classifier with the goal reduce the number of errors. This idea has been shown to improve deep neural's generalization capabilities in many computer vision tasks such as recognition and object localization. Apart from these applications, deep Neural Networks (CNNs) have also recently gained popularity in Time Series Classification (TSC) community. However, unlike in image problems, data augmentation techniques have not yet been thoroughly for the TSC task. This is surprising as the accuracy of learning models for TSC could potentially be improved, especially for datasets that exhibit overfitting, when a data augmentation method is. In this paper, we fill this gap by investigating the application of a proposed data augmentation technique based on the Dynamic Time Warping, for a deep learning model for TSC. To evaluate the potential of the training set, we performed extensive experiments using the UCR benchmark. Our preliminary experiments reveal that data augmentation can increase deep CNN's accuracy on some datasets and significantly the deep model's accuracy when the method is used in an ensemble.