Long Time-Series Data Augmentation via Timeseries Generative Adversarial Network

This study addresses the challenge of generating long-time-series data with high sampling rates, which is common in healthcare and semiconductors. We improve TimeGAN by replacing its GRU with an Independently Recurrent Neural Network (IndRNN), enhancing stability and performance on long sequences. Experiments show improved data augmentation and effective application in stock prediction, demonstrating the method's practicality for long time-series scenarios.