An comparative investigation into WGAN-GP, CTGAN, TimeGAN and DoppelGANger usage for generating synthetic time series finance data for use in forecasting model To recreate the experiment please run the Jupyter Notebook in order (s1-s7).
- s1_Data prep.ipynb to create the dataset
- s2_LSTM baseline model.ipynb to create the baseline prediction model
- s3_WGAN-GP.ipynb to create a synthetic dataset using WGAN-GP
- s4_CTGAN.ipynb to create a synthetic dataset using CTGAN and
- s5_TimeGAN.ipynb to create a synthetic dataset using TimeGAN
- s6_DopGAN.ipynb to create a synthetic dataset using DoppelGANger
- s7_Model_comparision to train and test the LSTM models trained on the synthetic and combination datasets & create the PCA and t-SNE plots
- indicator_generator.py is functions used for data preprocessing in s1_Data prep.ipynb