Time Series Data Imputation: Generative Adversarial Networks Approaches

Time series data analysis and forecasting is one of the key research areas in many industries. Incorrect or missing data causes a challenge in prediction and forecasting tasks. Hence, it is a necessity to have these portions filled in with most appropriate values in time series. With advancement in deep learning methods, there is an increased need for time series imputation. The traditional methods of imputation had many shortcomings such as – not suitable for large scale datasets, accuracy of results, difficulty in generating complex distributed data. Generative Adversarial Networks (GAN) has been one of the forefront mechanisms used to predict the missing data in areas like computer vision, natural language processing and more recently time series data sets. This session will focus on the imputation techniques for time series data which use the Generative Adversarial Networks (GAN) and compare their performances.