A Review on Time Series Data Augmentation Techniques for Deep Learning
Nor Adni Mat Leh, Mohammad Nizam Ibrahim, Muhammad Khusairi Osman, Anuar Mohamad, Mohd Najib Mohd Hussain · Journal of Advanced Research in Applied Sciences and Engineering Technology · 2024
Recently, deep artificial neural networks have gained huge attention in pattern recognition and classification. The classification models that are trained with insufficiently large datasets can contribute to degradation of generalization ability and the overall performance of the model. Therefore, the data augmentation technique can be applied to overcome the problems with limited dataset. Data augmentation is particularly popular with image data. On the other hand, data augmentation is less widely used for time series data to perform tasks such as classification and prediction. The augmentation method of time series dataset is the process of adding the size of dataset to be used to train the model and enhance the model performance. In this article, a review of augmentation methods for time series dataset and their applications to classification and forecasting is presented. The review shows that the GANs networks can generate synthetic data for load profiles. The conclusion drawn in this review article aims to provide a significant recommendation to help the researchers in selecting an appropriate approach to time series data augmentation especially for deep learning applications.